src/PVE/LXC.pm

   1 package PVE::LXC;
   2
   3 use strict;
   4 use warnings;
   5
   6 use POSIX qw(EINTR);
   7
   8 use Socket;
   9
  10 use File::Path;
  11 use File::Spec;
  12 use Cwd qw();
  13 use Fcntl qw(O_RDONLY O_WRONLY O_NOFOLLOW O_DIRECTORY);
  14 use Errno qw(ELOOP ENOTDIR EROFS ECONNREFUSED ENOSYS EEXIST);
  15 use IO::Socket::UNIX;
  16
  17 use PVE::Exception qw(raise_perm_exc);
  18 use PVE::Storage;
  19 use PVE::SafeSyslog;
  20 use PVE::INotify;
  21 use PVE::JSONSchema qw(get_standard_option);
  22 use PVE::Tools qw(
  23     run_command
  24     dir_glob_foreach
  25     file_get_contents
  26     file_set_contents
  27     AT_FDCWD
  28     O_PATH
  29     $IPV4RE
  30     $IPV6RE
  31 );
  32 use PVE::CpuSet;
  33 use PVE::Network;
  34 use PVE::AccessControl;
  35 use PVE::ProcFSTools;
  36 use PVE::Syscall qw(:fsmount);
  37 use PVE::LXC::Config;
  38 use PVE::GuestHelpers qw(safe_string_ne safe_num_ne safe_boolean_ne);
  39 use PVE::LXC::Tools;
  40 use PVE::LXC::CGroup;
  41 use PVE::LXC::Monitor;
  42 use PVE::CGroup;
  43
  44 use Time::HiRes qw (gettimeofday);
  45 my $have_sdn;
  46 eval {
  47     require PVE::Network::SDN::Zones;
  48     $have_sdn = 1;
  49 };
  50
  51 my $LXC_CONFIG_PATH = '/usr/share/lxc/config';
  52
  53 my $nodename = PVE::INotify::nodename();
  54
  55 my $cpuinfo= PVE::ProcFSTools::read_cpuinfo();
  56
  57 sub config_list {
  58     my $vmlist = PVE::Cluster::get_vmlist();
  59     my $res = {};
  60     return $res if !$vmlist || !$vmlist->{ids};
  61     my $ids = $vmlist->{ids};
  62
  63     foreach my $vmid (keys %$ids) {
  64         next if !$vmid; # skip CT0
  65         my $d = $ids->{$vmid};
  66         next if !$d->{node} || $d->{node} ne $nodename;
  67         next if !$d->{type} || $d->{type} ne 'lxc';
  68         $res->{$vmid} = { type => 'lxc', vmid => $vmid };
  69     }
  70     return $res;
  71 }
  72
  73 # container status helpers
  74
  75 sub list_active_containers {
  76
  77     my $filename = "/proc/net/unix";
  78
  79     # similar test is used by lcxcontainers.c: list_active_containers
  80     my $res = {};
  81
  82     my $fh = IO::File->new ($filename, "r");
  83     return $res if !$fh;
  84
  85     while (defined(my $line = <$fh>)) {
  86         if ($line =~ m/^[a-f0-9]+:\s+\S+\s+\S+\s+\S+\s+\S+\s+\S+\s+\d+\s+(\S+)$/) {
  87             my $path = $1;
  88             if ($path =~ m!^@/var/lib/lxc/(\d+)/command$!) {
  89                 $res->{$1} = 1;
  90             }
  91         }
  92     }
  93
  94     close($fh);
  95
  96     return $res;
  97 }
  98
  99 # warning: this is slow
 100 sub check_running {
 101     my ($vmid) = @_;
 102
 103     my $active_hash = list_active_containers();
 104
 105     return 1 if defined($active_hash->{$vmid});
 106
 107     return undef;
 108 }
 109
 110 sub get_container_disk_usage {
 111     my ($vmid, $pid) = @_;
 112
 113     return PVE::Tools::df("/proc/$pid/root/", 1);
 114 }
 115
 116 my $last_proc_vmid_stat;
 117
 118 our $vmstatus_return_properties = {
 119     vmid => get_standard_option('pve-vmid'),
 120     status => {
 121         description => "LXC Container status.",
 122         type => 'string',
 123         enum => ['stopped', 'running'],
 124     },
 125     maxmem => {
 126         description => "Maximum memory in bytes.",
 127         type => 'integer',
 128         optional => 1,
 129         renderer => 'bytes',
 130     },
 131     maxswap => {
 132         description => "Maximum SWAP memory in bytes.",
 133         type => 'integer',
 134         optional => 1,
 135         renderer => 'bytes',
 136     },
 137     maxdisk => {
 138         description => "Root disk size in bytes.",
 139         type => 'integer',
 140         optional => 1,
 141         renderer => 'bytes',
 142     },
 143     name => {
 144         description => "Container name.",
 145         type => 'string',
 146         optional => 1,
 147     },
 148     uptime => {
 149         description => "Uptime.",
 150         type => 'integer',
 151         optional => 1,
 152         renderer => 'duration',
 153     },
 154     cpus => {
 155         description => "Maximum usable CPUs.",
 156         type => 'number',
 157         optional => 1,
 158     },
 159     lock => {
 160         description => "The current config lock, if any.",
 161         type => 'string',
 162         optional => 1,
 163     },
 164     tags => {
 165         description => "The current configured tags, if any.",
 166         type => 'string',
 167         optional => 1,
 168     }
 169 };
 170
 171 sub vmstatus {
 172     my ($opt_vmid) = @_;
 173
 174     my $list = $opt_vmid ? { $opt_vmid => { type => 'lxc', vmid => $opt_vmid }} : config_list();
 175
 176     my $active_hash = list_active_containers();
 177
 178     my $cpucount = $cpuinfo->{cpus} || 1;
 179
 180     my $cdtime = gettimeofday;
 181
 182     my $uptime = (PVE::ProcFSTools::read_proc_uptime(1))[0];
 183     my $clock_ticks = POSIX::sysconf(&POSIX::_SC_CLK_TCK);
 184
 185     my $unprivileged = {};
 186
 187     foreach my $vmid (keys %$list) {
 188         my $d = $list->{$vmid};
 189
 190         eval { $d->{pid} = find_lxc_pid($vmid) if defined($active_hash->{$vmid}); };
 191         warn $@ if $@; # ignore errors (consider them stopped)
 192
 193         $d->{status} = $active_hash->{$vmid} ? 'running' : 'stopped';
 194
 195         my $cfspath = PVE::LXC::Config->cfs_config_path($vmid);
 196         my $conf = PVE::Cluster::cfs_read_file($cfspath) || {};
 197
 198         $unprivileged->{$vmid} = $conf->{unprivileged};
 199
 200         $d->{name} = $conf->{'hostname'} || "CT$vmid";
 201         $d->{name} =~ s/[\s]//g;
 202
 203         $d->{cpus} = $conf->{cores} || $conf->{cpulimit};
 204         $d->{cpus} = $cpucount if !$d->{cpus};
 205
 206         $d->{tags} = $conf->{tags} if defined($conf->{tags});
 207
 208         if ($d->{pid}) {
 209             my $res = get_container_disk_usage($vmid, $d->{pid});
 210             $d->{disk} = $res->{used};
 211             $d->{maxdisk} = $res->{total};
 212         } else {
 213             $d->{disk} = 0;
 214             # use 4GB by default ??
 215             if (my $rootfs = $conf->{rootfs}) {
 216                 my $rootinfo = PVE::LXC::Config->parse_volume('rootfs', $rootfs);
 217                 $d->{maxdisk} = $rootinfo->{size} || (4*1024*1024*1024);
 218             } else {
 219                 $d->{maxdisk} = 4*1024*1024*1024;
 220             }
 221         }
 222
 223         $d->{mem} = 0;
 224         $d->{swap} = 0;
 225         $d->{maxmem} = ($conf->{memory}||512)*1024*1024;
 226         $d->{maxswap} = ($conf->{swap}//0)*1024*1024;
 227
 228         $d->{uptime} = 0;
 229         $d->{cpu} = 0;
 230
 231         $d->{netout} = 0;
 232         $d->{netin} = 0;
 233
 234         $d->{diskread} = 0;
 235         $d->{diskwrite} = 0;
 236
 237         $d->{template} = 1 if PVE::LXC::Config->is_template($conf);
 238         $d->{lock} = $conf->{lock} if $conf->{lock};
 239     }
 240
 241     foreach my $vmid (keys %$list) {
 242         my $d = $list->{$vmid};
 243         my $pid = $d->{pid};
 244
 245         next if !$pid; # skip stopped CTs
 246
 247         my $proc_pid_stat = PVE::ProcFSTools::read_proc_pid_stat($pid);
 248         $d->{uptime} = int(($uptime - $proc_pid_stat->{starttime}) / $clock_ticks); # the method lxcfs uses
 249
 250         my $unpriv = $unprivileged->{$vmid};
 251
 252         my $cgroups = PVE::LXC::CGroup->new($vmid);
 253
 254         if (defined(my $mem = $cgroups->get_memory_stat())) {
 255             $d->{mem} = $mem->{mem};
 256             $d->{swap} = $mem->{swap};
 257         } else {
 258             $d->{mem} = 0;
 259             $d->{swap} = 0;
 260         }
 261
 262         if (defined(my $blkio = $cgroups->get_io_stats())) {
 263             $d->{diskread} = $blkio->{diskread};
 264             $d->{diskwrite} = $blkio->{diskwrite};
 265         } else {
 266             $d->{diskread} = 0;
 267             $d->{diskwrite} = 0;
 268         }
 269
 270         if (defined(my $cpu = $cgroups->get_cpu_stat())) {
 271             # Total time (in milliseconds) used up by the cpu.
 272             my $used_ms = $cpu->{utime} + $cpu->{stime};
 273
 274             my $old = $last_proc_vmid_stat->{$vmid};
 275             if (!$old) {
 276                 $last_proc_vmid_stat->{$vmid} = {
 277                     time => $cdtime,
 278                     used => $used_ms,
 279                     cpu => 0,
 280                 };
 281                 next;
 282             }
 283
 284             my $delta_ms = ($cdtime - $old->{time}) * $cpucount * 1000.0;
 285             if ($delta_ms > 1000.0) {
 286                 my $delta_used_ms = $used_ms - $old->{used};
 287                 $d->{cpu} = (($delta_used_ms / $delta_ms) * $cpucount) / $d->{cpus};
 288                 $last_proc_vmid_stat->{$vmid} = {
 289                     time => $cdtime,
 290                     used => $used_ms,
 291                     cpu => $d->{cpu},
 292                 };
 293             } else {
 294                 $d->{cpu} = $old->{cpu};
 295             }
 296         } else {
 297             $d->{cpu} = 0;
 298         }
 299     }
 300
 301     my $netdev = PVE::ProcFSTools::read_proc_net_dev();
 302
 303     foreach my $dev (keys %$netdev) {
 304         next if $dev !~ m/^veth([1-9]\d*)i/;
 305         my $vmid = $1;
 306         my $d = $list->{$vmid};
 307
 308         next if !$d;
 309
 310         $d->{netout} += $netdev->{$dev}->{receive};
 311         $d->{netin} += $netdev->{$dev}->{transmit};
 312
 313     }
 314
 315     return $list;
 316 }
 317
 318 sub find_lxc_console_pids {
 319
 320     my $res = {};
 321
 322     PVE::Tools::dir_glob_foreach('/proc', '\d+', sub {
 323         my ($pid) = @_;
 324
 325         my $cmdline = PVE::Tools::file_read_firstline("/proc/$pid/cmdline");
 326         return if !$cmdline;
 327
 328         my @args = split(/\0/, $cmdline);
 329
 330         # search for lxc-console -n <vmid>
 331         return if scalar(@args) != 3;
 332         return if $args[1] ne '-n';
 333         return if $args[2] !~ m/^\d+$/;
 334         return if $args[0] !~ m|^(/usr/bin/)?lxc-console$|;
 335
 336         my $vmid = $args[2];
 337
 338         push @{$res->{$vmid}}, $pid;
 339     });
 340
 341     return $res;
 342 }
 343
 344 sub find_lxc_pid {
 345     my ($vmid) = @_;
 346
 347     my $pid = undef;
 348     my $parser = sub {
 349         my $line = shift;
 350         $pid = $1 if $line =~ m/^PID:\s+(\d+)$/;
 351     };
 352     PVE::Tools::run_command(['lxc-info', '-n', $vmid, '-p'], outfunc => $parser);
 353
 354     die "unable to get PID for CT $vmid (not running?)\n" if !$pid;
 355
 356     return $pid;
 357 }
 358
 359 sub open_pid_fd($) {
 360     my ($pid) = @_;
 361     sysopen(my $fd, "/proc/$pid", O_RDONLY | O_DIRECTORY)
 362         or die "failed to open /proc/$pid pid fd\n";
 363     return $fd;
 364 }
 365
 366 sub open_lxc_pid {
 367     my ($vmid) = @_;
 368
 369     # Find the pid and open:
 370     my $pid = find_lxc_pid($vmid);
 371     my $fd = open_pid_fd($pid);
 372
 373     # Verify:
 374     my $pid2 = find_lxc_pid($vmid);
 375
 376     return () if $pid != $pid2;
 377     return ($pid, $fd);
 378 }
 379
 380 sub open_ppid {
 381     my ($pid) = @_;
 382
 383     # Find the parent pid via proc and open it:
 384     my $stat = PVE::ProcFSTools::read_proc_pid_stat($pid);
 385     my $ppid = $stat->{ppid} // die "failed to get parent pid\n";
 386
 387     my $fd = open_pid_fd($ppid);
 388
 389     # Verify:
 390     $stat = PVE::ProcFSTools::read_proc_pid_stat($pid);
 391     my $ppid2 = $stat->{ppid} // die "failed to get parent pid for verification\n";
 392
 393     return () if $ppid != $ppid2;
 394     return ($ppid, $fd);
 395 }
 396
 397 # Note: we cannot use Net:IP, because that only allows strict
 398 # CIDR networks
 399 sub parse_ipv4_cidr {
 400     my ($cidr, $noerr) = @_;
 401
 402     if ($cidr =~ m!^($IPV4RE)(?:/(\d+))$! && ($2 > 7) &&  ($2 <= 32)) {
 403         return { address => $1, netmask => $PVE::Network::ipv4_reverse_mask->[$2] };
 404     }
 405
 406     return undef if $noerr;
 407
 408     die "unable to parse ipv4 address/mask\n";
 409 }
 410
 411 # Deprecated. Use `PVE::CGroup::get_cgroup_controllers()` instead.
 412 sub get_cgroup_subsystems {
 413     PVE::CGroup::get_v1_controllers();
 414 }
 415
 416 # With seccomp trap to userspace we now have the ability to optionally forward
 417 # certain syscalls to the "host" to handle (via our pve-lxc-syscalld daemon).
 418 #
 419 # This means that there are cases where we need to create an extra seccomp
 420 # profile for the container to load.
 421 #
 422 # This returns a configuration snippet added to the raw lxc config.
 423 sub make_seccomp_config {
 424     my ($conf, $vmid, $conf_dir, $unprivileged, $features) = @_;
 425     # User-configured profile has precedence, note that the user's entry would
 426     # be written 'after' this line anyway...
 427     if (PVE::LXC::Config->has_lxc_entry($conf, 'lxc.seccomp.profile')) {
 428         # Warn the user if this conflicts with a feature:
 429         my $warn = join(', ', grep { $features->{$_} } qw(keyctl mknod));
 430         warn "explicitly configured lxc.seccomp.profile overrides the following settings: $warn\n"
 431             if length($warn) > 0;
 432         return '';
 433     }
 434
 435     # Privileged containers keep using the default (which is already part of
 436     # the files included via lxc.include, so we don't need to write it out,
 437     # that way it stays admin-configurable via /usr/share/lxc/config/... as
 438     # well)
 439     return '' if !$unprivileged;
 440
 441     my $rules = {
 442         keyctl => ['errno 38'],
 443     };
 444
 445     my $raw_conf = '';
 446
 447     # Unprivileged containers will get keyctl() disabled by default as a
 448     # workaround for systemd-networkd behavior. But we have an option to
 449     # explicitly enable it:
 450     if ($features->{keyctl}) {
 451         delete $rules->{keyctl};
 452     }
 453
 454     # By default, unprivileged containers cannot use `mknod` at all.
 455     # Since lxc 3.2, we can use seccomp's trap to userspace feature for this,
 456     # but for now this is experimental, so it has to be enabled via a feature
 457     # flag.
 458     # Note that we only handle block and char devices (like lxd), the rest we
 459     # leave up to the kernel. We may in the future remove this if seccomp gets
 460     # a way to tell the kernel to "continue" a syscall.
 461     if ($features->{mknod}) {
 462         my ($ok, $kernel) = PVE::ProcFSTools::check_kernel_release(5, 3);
 463         if (!$ok) {
 464             die "'mknod' feature requested, but kernel too old (found $kernel, required >= 5.3)\n";
 465         }
 466
 467         $raw_conf .= "lxc.seccomp.notify.proxy = unix:/run/pve/lxc-syscalld.sock\n";
 468         $raw_conf .= "lxc.seccomp.notify.cookie = $vmid\n";
 469
 470         $rules->{mknod} = [
 471             # condition: (mode & S_IFMT) == S_IFCHR
 472             'notify [1,8192,SCMP_CMP_MASKED_EQ,61440]',
 473             # condition: (mode & S_IFMT) == S_IFBLK
 474             'notify [1,24576,SCMP_CMP_MASKED_EQ,61440]',
 475         ];
 476         $rules->{mknodat} = [
 477             # condition: (mode & S_IFMT) == S_IFCHR
 478             'notify [2,8192,SCMP_CMP_MASKED_EQ,61440]',
 479             # condition: (mode & S_IFMT) == S_IFBLK
 480             'notify [2,24576,SCMP_CMP_MASKED_EQ,61440]',
 481         ];
 482     }
 483
 484     # Now build the custom seccomp rule text...
 485     my $extra_rules = join("\n", map {
 486         my $syscall = $_;
 487         map { "$syscall $_" } $rules->{$syscall}->@*
 488     } sort keys %$rules) . "\n";
 489
 490     return $raw_conf if $extra_rules eq "\n";
 491
 492     # We still have the "most common" config readily available, so don't write
 493     # out that one:
 494     if ($raw_conf eq '' && $extra_rules eq "keyctl errno 38\n") {
 495         # we have no extra $raw_conf and use the same we had in pve 6.1:
 496         return "lxc.seccomp.profile = $LXC_CONFIG_PATH/pve-userns.seccomp\n";
 497     }
 498
 499     # Write the rule file to the container's config path:
 500     my $rule_file = "$conf_dir/rules.seccomp";
 501     my $rule_data = file_get_contents("$LXC_CONFIG_PATH/common.seccomp")
 502         . $extra_rules;
 503     file_set_contents($rule_file, $rule_data);
 504     $raw_conf .= "lxc.seccomp.profile = $rule_file\n";
 505
 506     return $raw_conf;
 507 }
 508
 509 # Since lxc-3.0.2 we can have lxc generate a profile for the container
 510 # automatically. The default should be equivalent to the old
 511 # `lxc-container-default-cgns` profile.
 512 #
 513 # Additionally this also added `lxc.apparmor.raw` which can be used to inject
 514 # additional lines into the profile. We can use that to allow mounting specific
 515 # file systems.
 516 sub make_apparmor_config {
 517     my ($conf, $unprivileged, $features) = @_;
 518
 519     # user-configured profile has precedence, but first we go through our own
 520     # code to figure out whether we should warn the user:
 521
 522     my $raw = "lxc.apparmor.profile = generated\n";
 523     my @profile_uses;
 524
 525     if ($features->{fuse}) {
 526         # For the informational warning:
 527         push @profile_uses, 'features:fuse';
 528     }
 529
 530     # There's lxc.apparmor.allow_nesting now, which will add the necessary
 531     # apparmor lines, create an apparmor namespace for the container, but also
 532     # adds proc and sysfs mounts to /dev/.lxc/{proc,sys}. These do not have
 533     # lxcfs mounted over them, because that would prevent the container from
 534     # mounting new instances of them for nested containers.
 535     if ($features->{nesting}) {
 536         push @profile_uses, 'features:nesting';
 537         $raw .= "lxc.apparmor.allow_nesting = 1\n"
 538     } else {
 539         # In the default profile in /etc/apparmor.d we patch this in because
 540         # otherwise a container can for example run `chown` on /sys, breaking
 541         # access to it for non-CAP_DAC_OVERRIDE tools on the host:
 542         $raw .= "lxc.apparmor.raw = deny mount -> /proc/,\n";
 543         $raw .= "lxc.apparmor.raw = deny mount -> /sys/,\n";
 544         # Preferably we could use the 'remount' flag but this does not sit well
 545         # with apparmor_parser currently:
 546         #  mount options=(rw, nosuid, nodev, noexec, remount) -> /sys/,
 547     }
 548
 549     if (my $mount = $features->{mount}) {
 550         push @profile_uses, 'features:mount';
 551         foreach my $fs (PVE::Tools::split_list($mount)) {
 552             $raw .= "lxc.apparmor.raw = mount fstype=$fs,\n";
 553         }
 554     }
 555
 556     # More to come?
 557
 558     if (PVE::LXC::Config->has_lxc_entry($conf, 'lxc.apparmor.profile')) {
 559         if (length(my $used = join(', ', @profile_uses))) {
 560             warn "explicitly configured lxc.apparmor.profile overrides the following settings: $used\n";
 561         }
 562         return '';
 563     }
 564
 565     return $raw;
 566 }
 567
 568 sub update_lxc_config {
 569     my ($vmid, $conf) = @_;
 570
 571     my $dir = "/var/lib/lxc/$vmid";
 572
 573     if ($conf->{template}) {
 574
 575         unlink "$dir/config";
 576
 577         return;
 578     }
 579
 580     my ($lxc_major, $lxc_minor) = get_lxc_version();
 581
 582     my $raw = '';
 583
 584     if ($lxc_major >= 4) {
 585         # Explicitly don't use relative directories, which is the default, but
 586         # note that we do this mostly because they are only applied for *some*
 587         # cgroups. Our pve-container@.service now starts lxc-start with `-F`,
 588         # so we also don't need to worry about the new monitor cgroup to
 589         # confuse systemd.
 590         $raw .= "lxc.cgroup.relative = 0\n";
 591
 592         # To make things easier, let's keep our previous cgroup layout and
 593         # simply move the monitor outside:
 594         $raw .= "lxc.cgroup.dir.monitor = lxc.monitor/$vmid\n";
 595         # cgroup namespace separation for stronger limits:
 596         $raw .= "lxc.cgroup.dir.container = lxc/$vmid\n";
 597         $raw .= "lxc.cgroup.dir.container.inner = ns\n";
 598     }
 599
 600     die "missing 'arch' - internal error" if !$conf->{arch};
 601     $raw .= "lxc.arch = $conf->{arch}\n";
 602
 603     my $custom_idmap = PVE::LXC::Config->has_lxc_entry($conf, 'lxc.idmap');
 604     my $unprivileged = $conf->{unprivileged} || $custom_idmap;
 605
 606     my $ostype = $conf->{ostype} || die "missing 'ostype' - internal error";
 607
 608     my $cfgpath = '/usr/share/lxc/config';
 609     my $inc = "$cfgpath/$ostype.common.conf";
 610     $inc ="$cfgpath/common.conf" if !-f $inc;
 611     $raw .= "lxc.include = $inc\n";
 612     if ($unprivileged) {
 613         $inc = "$cfgpath/$ostype.userns.conf";
 614         $inc = "$cfgpath/userns.conf" if !-f $inc;
 615         $raw .= "lxc.include = $inc\n";
 616     }
 617
 618     my $features = PVE::LXC::Config->parse_features($conf->{features});
 619
 620     $raw .= make_seccomp_config($conf, $vmid, $dir, $unprivileged, $features);
 621     $raw .= make_apparmor_config($conf, $unprivileged, $features);
 622     if ($features->{fuse}) {
 623         $raw .= "lxc.apparmor.raw = mount fstype=fuse,\n";
 624         $raw .= "lxc.mount.entry = /dev/fuse dev/fuse none bind,create=file 0 0\n";
 625     }
 626
 627     if ($unprivileged && !$features->{force_rw_sys}) {
 628         # unpriv. CT default to sys:rw, but that doesn't always plays well with
 629         # systemd, e.g., systemd-networkd https://systemd.io/CONTAINER_INTERFACE/
 630         $raw .= "lxc.mount.auto = sys:mixed\n";
 631     }
 632
 633     # WARNING: DO NOT REMOVE this without making sure that loop device nodes
 634     # cannot be exposed to the container with r/w access (cgroup perms).
 635     # When this is enabled mounts will still remain in the monitor's namespace
 636     # after the container unmounted them and thus will not detach from their
 637     # files while the container is running!
 638     $raw .= "lxc.monitor.unshare = 1\n";
 639
 640     my $cgv1 = get_cgroup_subsystems();
 641
 642     # Should we read them from /etc/subuid?
 643     if ($unprivileged && !$custom_idmap) {
 644         $raw .= "lxc.idmap = u 0 100000 65536\n";
 645         $raw .= "lxc.idmap = g 0 100000 65536\n";
 646     }
 647
 648     if (!PVE::LXC::Config->has_dev_console($conf)) {
 649         $raw .= "lxc.console.path = none\n";
 650         $raw .= "lxc.cgroup.devices.deny = c 5:1 rwm\n" if $cgv1->{devices};
 651     }
 652
 653     my $ttycount = PVE::LXC::Config->get_tty_count($conf);
 654     $raw .= "lxc.tty.max = $ttycount\n";
 655
 656     # some init scripts expect a linux terminal (turnkey).
 657     $raw .= "lxc.environment = TERM=linux\n";
 658
 659     my $utsname = $conf->{hostname} || "CT$vmid";
 660     $raw .= "lxc.uts.name = $utsname\n";
 661
 662     if ($cgv1->{memory}) {
 663         my $memory = $conf->{memory} || 512;
 664         my $swap = $conf->{swap} // 0;
 665
 666         my $lxcmem = int($memory*1024*1024);
 667         $raw .= "lxc.cgroup.memory.limit_in_bytes = $lxcmem\n";
 668
 669         my $lxcswap = int(($memory + $swap)*1024*1024);
 670         $raw .= "lxc.cgroup.memory.memsw.limit_in_bytes = $lxcswap\n";
 671     }
 672
 673     if ($cgv1->{cpu}) {
 674         if (my $cpulimit = $conf->{cpulimit}) {
 675             $raw .= "lxc.cgroup.cpu.cfs_period_us = 100000\n";
 676             my $value = int(100000*$cpulimit);
 677             $raw .= "lxc.cgroup.cpu.cfs_quota_us = $value\n";
 678         }
 679
 680         my $shares = $conf->{cpuunits} || 1024;
 681         $raw .= "lxc.cgroup.cpu.shares = $shares\n";
 682     }
 683
 684     die "missing 'rootfs' configuration\n"
 685         if !defined($conf->{rootfs});
 686
 687     my $mountpoint = PVE::LXC::Config->parse_volume('rootfs', $conf->{rootfs});
 688
 689     $raw .= "lxc.rootfs.path = $dir/rootfs\n";
 690
 691     foreach my $k (sort keys %$conf) {
 692         next if $k !~ m/^net(\d+)$/;
 693         my $ind = $1;
 694         my $d = PVE::LXC::Config->parse_lxc_network($conf->{$k});
 695         $raw .= "lxc.net.$ind.type = veth\n";
 696         $raw .= "lxc.net.$ind.veth.pair = veth${vmid}i${ind}\n";
 697         $raw .= "lxc.net.$ind.hwaddr = $d->{hwaddr}\n" if defined($d->{hwaddr});
 698         $raw .= "lxc.net.$ind.name = $d->{name}\n" if defined($d->{name});
 699         $raw .= "lxc.net.$ind.mtu = $d->{mtu}\n" if defined($d->{mtu});
 700
 701         # Starting with lxc 4.0, we do not patch lxc to execute our up-scripts.
 702         if ($lxc_major >= 4) {
 703             $raw .= "lxc.net.$ind.script.up = /usr/share/lxc/lxcnetaddbr\n";
 704         }
 705     }
 706
 707     my $had_cpuset = 0;
 708     if (my $lxcconf = $conf->{lxc}) {
 709         foreach my $entry (@$lxcconf) {
 710             my ($k, $v) = @$entry;
 711             $had_cpuset = 1 if $k eq 'lxc.cgroup.cpuset.cpus' || $k eq 'lxc.cgroup2.cpuset.cpus';
 712             $raw .= "$k = $v\n";
 713         }
 714     }
 715
 716     my $cpuset;
 717     my ($cpuset_cgroup, $cpuset_version) = eval { PVE::CGroup::cpuset_controller_path() };
 718     if (defined($cpuset_cgroup)) {
 719         $cpuset = eval { PVE::CpuSet->new_from_path("$cpuset_cgroup/lxc", 1) }
 720             || PVE::CpuSet->new_from_path($cpuset_cgroup, 1);
 721     }
 722     my $cores = $conf->{cores};
 723     if (!$had_cpuset && $cores && $cpuset) {
 724         my @members = $cpuset->members();
 725         while (scalar(@members) > $cores) {
 726             my $randidx = int(rand(scalar(@members)));
 727             $cpuset->delete($members[$randidx]);
 728             splice(@members, $randidx, 1); # keep track of the changes
 729         }
 730         my $ver = $cpuset_version == 1 ? '' : '2';
 731         $raw .= "lxc.cgroup$ver.cpuset.cpus = ".$cpuset->short_string()."\n";
 732     }
 733
 734     File::Path::mkpath("$dir/rootfs");
 735
 736     PVE::Tools::file_set_contents("$dir/config", $raw);
 737 }
 738
 739 # verify and cleanup nameserver list (replace \0 with ' ')
 740 sub verify_nameserver_list {
 741     my ($nameserver_list) = @_;
 742
 743     my @list = ();
 744     foreach my $server (PVE::Tools::split_list($nameserver_list)) {
 745         PVE::LXC::Config::verify_ip_with_ll_iface($server);
 746         push @list, $server;
 747     }
 748
 749     return join(' ', @list);
 750 }
 751
 752 sub verify_searchdomain_list {
 753     my ($searchdomain_list) = @_;
 754
 755     my @list = ();
 756     foreach my $server (PVE::Tools::split_list($searchdomain_list)) {
 757         # todo: should we add checks for valid dns domains?
 758         push @list, $server;
 759     }
 760
 761     return join(' ', @list);
 762 }
 763
 764 sub get_console_command {
 765     my ($vmid, $conf, $escapechar) = @_;
 766
 767     # '-1' as $escapechar disables keyboard escape sequence
 768     # any other passed char (a-z) will result in <Ctrl+$escapechar q>
 769
 770     my $cmode = PVE::LXC::Config->get_cmode($conf);
 771
 772     my $cmd = [];
 773     if ($cmode eq 'console') {
 774         push @$cmd, 'lxc-console', '-n',  $vmid, '-t', 0;
 775         push @$cmd, '-e', $escapechar if $escapechar;
 776     } elsif ($cmode eq 'tty') {
 777         push @$cmd, 'lxc-console', '-n',  $vmid;
 778         push @$cmd, '-e', $escapechar if $escapechar;
 779     } elsif ($cmode eq 'shell') {
 780         push @$cmd, 'lxc-attach', '--clear-env', '-n', $vmid;
 781     } else {
 782         die "internal error";
 783     }
 784
 785     return $cmd;
 786 }
 787
 788 sub get_primary_ips {
 789     my ($conf) = @_;
 790
 791     # return data from net0
 792
 793     return undef if !defined($conf->{net0});
 794     my $net = PVE::LXC::Config->parse_lxc_network($conf->{net0});
 795
 796     my $ipv4 = $net->{ip};
 797     if ($ipv4) {
 798         if ($ipv4 =~ /^(dhcp|manual)$/) {
 799             $ipv4 = undef
 800         } else {
 801             $ipv4 =~ s!/\d+$!!;
 802         }
 803     }
 804     my $ipv6 = $net->{ip6};
 805     if ($ipv6) {
 806         if ($ipv6 =~ /^(auto|dhcp|manual)$/) {
 807             $ipv6 = undef;
 808         } else {
 809             $ipv6 =~ s!/\d+$!!;
 810         }
 811     }
 812
 813     return ($ipv4, $ipv6);
 814 }
 815
 816 sub delete_mountpoint_volume {
 817     my ($storage_cfg, $vmid, $volume) = @_;
 818
 819     return if PVE::LXC::Config->classify_mountpoint($volume) ne 'volume';
 820
 821     my ($vtype, $name, $owner) = PVE::Storage::parse_volname($storage_cfg, $volume);
 822
 823     if ($vmid == $owner) {
 824         PVE::Storage::vdisk_free($storage_cfg, $volume);
 825     } else {
 826         warn "ignore deletion of '$volume', CT $vmid isn't the owner!\n";
 827     }
 828 }
 829
 830 sub destroy_lxc_container {
 831     my ($storage_cfg, $vmid, $conf, $replacement_conf, $purge_unreferenced) = @_;
 832
 833     PVE::LXC::Config->foreach_volume_full($conf, {include_unused => 1}, sub {
 834         my ($ms, $mountpoint) = @_;
 835         delete_mountpoint_volume($storage_cfg, $vmid, $mountpoint->{volume});
 836     });
 837
 838     if ($purge_unreferenced) { # also remove unreferenced disk
 839         my $vmdisks = PVE::Storage::vdisk_list($storage_cfg, undef, $vmid, undef, 'rootdir');
 840         PVE::Storage::foreach_volid($vmdisks, sub {
 841             my ($volid, $sid, $volname, $d) = @_;
 842             eval { PVE::Storage::vdisk_free($storage_cfg, $volid) };
 843             warn $@ if $@;
 844         });
 845     }
 846
 847     rmdir "/var/lib/lxc/$vmid/rootfs";
 848     unlink "/var/lib/lxc/$vmid/config";
 849     rmdir "/var/lib/lxc/$vmid";
 850     if (defined $replacement_conf) {
 851         PVE::LXC::Config->write_config($vmid, $replacement_conf);
 852     } else {
 853         PVE::LXC::Config->destroy_config($vmid);
 854     }
 855 }
 856
 857 sub vm_stop_cleanup {
 858     my ($storage_cfg, $vmid, $conf, $keepActive) = @_;
 859
 860     return if $keepActive;
 861
 862     eval {
 863         my $vollist = PVE::LXC::Config->get_vm_volumes($conf);
 864         PVE::Storage::deactivate_volumes($storage_cfg, $vollist);
 865     };
 866     warn $@ if $@; # avoid errors - just warn
 867 }
 868
 869 sub update_net {
 870     my ($vmid, $conf, $opt, $newnet, $netid, $rootdir) = @_;
 871
 872     if ($newnet->{type} ne 'veth') {
 873         # for when there are physical interfaces
 874         die "cannot update interface of type $newnet->{type}";
 875     }
 876
 877     my $veth = "veth${vmid}i${netid}";
 878     my $eth = $newnet->{name};
 879
 880     if (my $oldnetcfg = $conf->{$opt}) {
 881         my $oldnet = PVE::LXC::Config->parse_lxc_network($oldnetcfg);
 882
 883         if (safe_string_ne($oldnet->{hwaddr}, $newnet->{hwaddr}) ||
 884             safe_string_ne($oldnet->{name}, $newnet->{name})) {
 885
 886             PVE::Network::veth_delete($veth);
 887             delete $conf->{$opt};
 888             PVE::LXC::Config->write_config($vmid, $conf);
 889
 890             hotplug_net($vmid, $conf, $opt, $newnet, $netid);
 891
 892         } else {
 893             if (safe_string_ne($oldnet->{bridge}, $newnet->{bridge}) ||
 894                 safe_num_ne($oldnet->{tag}, $newnet->{tag}) ||
 895                 safe_num_ne($oldnet->{firewall}, $newnet->{firewall})) {
 896
 897                 if ($oldnet->{bridge}) {
 898                     PVE::Network::tap_unplug($veth);
 899                     foreach (qw(bridge tag firewall)) {
 900                         delete $oldnet->{$_};
 901                     }
 902                     $conf->{$opt} = PVE::LXC::Config->print_lxc_network($oldnet);
 903                     PVE::LXC::Config->write_config($vmid, $conf);
 904                 }
 905
 906                 if ($have_sdn) {
 907                     PVE::Network::SDN::Zones::tap_plug($veth, $newnet->{bridge}, $newnet->{tag}, $newnet->{firewall}, $newnet->{trunks}, $newnet->{rate});
 908                 } else {
 909                     PVE::Network::tap_plug($veth, $newnet->{bridge}, $newnet->{tag}, $newnet->{firewall}, $newnet->{trunks}, $newnet->{rate});
 910                 }
 911
 912                 # This includes the rate:
 913                 foreach (qw(bridge tag firewall rate)) {
 914                     $oldnet->{$_} = $newnet->{$_} if $newnet->{$_};
 915                 }
 916             } elsif (safe_string_ne($oldnet->{rate}, $newnet->{rate})) {
 917                 # Rate can be applied on its own but any change above needs to
 918                 # include the rate in tap_plug since OVS resets everything.
 919                 PVE::Network::tap_rate_limit($veth, $newnet->{rate});
 920                 $oldnet->{rate} = $newnet->{rate}
 921             }
 922             $conf->{$opt} = PVE::LXC::Config->print_lxc_network($oldnet);
 923             PVE::LXC::Config->write_config($vmid, $conf);
 924         }
 925     } else {
 926         hotplug_net($vmid, $conf, $opt, $newnet, $netid);
 927     }
 928
 929     update_ipconfig($vmid, $conf, $opt, $eth, $newnet, $rootdir);
 930 }
 931
 932 sub hotplug_net {
 933     my ($vmid, $conf, $opt, $newnet, $netid) = @_;
 934
 935     my $veth = "veth${vmid}i${netid}";
 936     my $vethpeer = $veth . "p";
 937     my $eth = $newnet->{name};
 938
 939     if ($have_sdn) {
 940         PVE::Network::SDN::Zones::veth_create($veth, $vethpeer, $newnet->{bridge}, $newnet->{hwaddr});
 941         PVE::Network::SDN::Zones::tap_plug($veth, $newnet->{bridge}, $newnet->{tag}, $newnet->{firewall}, $newnet->{trunks}, $newnet->{rate});
 942     } else {
 943         PVE::Network::veth_create($veth, $vethpeer, $newnet->{bridge}, $newnet->{hwaddr});
 944         PVE::Network::tap_plug($veth, $newnet->{bridge}, $newnet->{tag}, $newnet->{firewall}, $newnet->{trunks}, $newnet->{rate});
 945     }
 946
 947     # attach peer in container
 948     my $cmd = ['lxc-device', '-n', $vmid, 'add', $vethpeer, "$eth" ];
 949     PVE::Tools::run_command($cmd);
 950
 951     # link up peer in container
 952     $cmd = ['lxc-attach', '-n', $vmid, '-s', 'NETWORK', '--', '/sbin/ip', 'link', 'set', $eth ,'up'  ];
 953     PVE::Tools::run_command($cmd);
 954
 955     my $done = { type => 'veth' };
 956     foreach (qw(bridge tag firewall hwaddr name)) {
 957         $done->{$_} = $newnet->{$_} if $newnet->{$_};
 958     }
 959     $conf->{$opt} = PVE::LXC::Config->print_lxc_network($done);
 960
 961     PVE::LXC::Config->write_config($vmid, $conf);
 962 }
 963
 964 sub update_ipconfig {
 965     my ($vmid, $conf, $opt, $eth, $newnet, $rootdir) = @_;
 966
 967     my $lxc_setup = PVE::LXC::Setup->new($conf, $rootdir);
 968
 969     my $optdata = PVE::LXC::Config->parse_lxc_network($conf->{$opt});
 970     my $deleted = [];
 971     my $added = [];
 972     my $nscmd = sub {
 973         my $cmdargs = shift;
 974         PVE::Tools::run_command(['lxc-attach', '-n', $vmid, '-s', 'NETWORK', '--', @_], %$cmdargs);
 975     };
 976     my $ipcmd = sub { &$nscmd({}, '/sbin/ip', @_) };
 977
 978     my $change_ip_config = sub {
 979         my ($ipversion) = @_;
 980
 981         my $family_opt = "-$ipversion";
 982         my $suffix = $ipversion == 4 ? '' : $ipversion;
 983         my $gw= "gw$suffix";
 984         my $ip= "ip$suffix";
 985
 986         my $newip = $newnet->{$ip};
 987         my $newgw = $newnet->{$gw};
 988         my $oldip = $optdata->{$ip};
 989         my $oldgw = $optdata->{$gw};
 990
 991         my $change_ip = safe_string_ne($oldip, $newip);
 992         my $change_gw = safe_string_ne($oldgw, $newgw);
 993
 994         return if !$change_ip && !$change_gw;
 995
 996         # step 1: add new IP, if this fails we cancel
 997         my $is_real_ip = ($newip && $newip !~ /^(?:auto|dhcp|manual)$/);
 998         if ($change_ip && $is_real_ip) {
 999             eval { &$ipcmd($family_opt, 'addr', 'add', $newip, 'dev', $eth); };
1000             if (my $err = $@) {
1001                 warn $err;
1002                 return;
1003             }
1004         }
1005
1006         # step 2: replace gateway
1007         #   If this fails we delete the added IP and cancel.
1008         #   If it succeeds we save the config and delete the old IP, ignoring
1009         #   errors. The config is then saved.
1010         # Note: 'ip route replace' can add
1011         if ($change_gw) {
1012             if ($newgw) {
1013                 eval {
1014                     if ($is_real_ip && !PVE::Network::is_ip_in_cidr($newgw, $newip, $ipversion)) {
1015                         &$ipcmd($family_opt, 'route', 'add', $newgw, 'dev', $eth);
1016                     }
1017                     &$ipcmd($family_opt, 'route', 'replace', 'default', 'via', $newgw);
1018                 };
1019                 if (my $err = $@) {
1020                     warn $err;
1021                     # the route was not replaced, the old IP is still available
1022                     # rollback (delete new IP) and cancel
1023                     if ($change_ip) {
1024                         eval { &$ipcmd($family_opt, 'addr', 'del', $newip, 'dev', $eth); };
1025                         warn $@ if $@; # no need to die here
1026                     }
1027                     return;
1028                 }
1029             } else {
1030                 eval { &$ipcmd($family_opt, 'route', 'del', 'default'); };
1031                 # if the route was not deleted, the guest might have deleted it manually
1032                 # warn and continue
1033                 warn $@ if $@;
1034             }
1035             if ($oldgw && $oldip && !PVE::Network::is_ip_in_cidr($oldgw, $oldip)) {
1036                 eval { &$ipcmd($family_opt, 'route', 'del', $oldgw, 'dev', $eth); };
1037                 # warn if the route was deleted manually
1038                 warn $@ if $@;
1039             }
1040         }
1041
1042         # from this point on we save the configuration
1043         # step 3: delete old IP ignoring errors
1044         if ($change_ip && $oldip && $oldip !~ /^(?:auto|dhcp)$/) {
1045             # We need to enable promote_secondaries, otherwise our newly added
1046             # address will be removed along with the old one.
1047             my $promote = 0;
1048             eval {
1049                 if ($ipversion == 4) {
1050                     &$nscmd({ outfunc => sub { $promote = int(shift) } },
1051                             'cat', "/proc/sys/net/ipv4/conf/$eth/promote_secondaries");
1052                     &$nscmd({}, 'sysctl', "net.ipv4.conf.$eth.promote_secondaries=1");
1053                 }
1054                 &$ipcmd($family_opt, 'addr', 'del', $oldip, 'dev', $eth);
1055             };
1056             warn $@ if $@; # no need to die here
1057
1058             if ($ipversion == 4) {
1059                 &$nscmd({}, 'sysctl', "net.ipv4.conf.$eth.promote_secondaries=$promote");
1060             }
1061         }
1062
1063         foreach my $property ($ip, $gw) {
1064             if ($newnet->{$property}) {
1065                 $optdata->{$property} = $newnet->{$property};
1066             } else {
1067                 delete $optdata->{$property};
1068             }
1069         }
1070         $conf->{$opt} = PVE::LXC::Config->print_lxc_network($optdata);
1071         PVE::LXC::Config->write_config($vmid, $conf);
1072         $lxc_setup->setup_network($conf);
1073     };
1074
1075     &$change_ip_config(4);
1076     &$change_ip_config(6);
1077
1078 }
1079
1080 my $open_namespace = sub {
1081     my ($vmid, $pid, $kind) = @_;
1082     sysopen my $fd, "/proc/$pid/ns/$kind", O_RDONLY
1083         or die "failed to open $kind namespace of container $vmid: $!\n";
1084     return $fd;
1085 };
1086
1087 my $enter_namespace = sub {
1088     my ($vmid, $pid, $kind, $type) = @_;
1089     my $fd = $open_namespace->($vmid, $pid, $kind);
1090     PVE::Tools::setns(fileno($fd), $type)
1091         or die "failed to enter $kind namespace of container $vmid: $!\n";
1092     close $fd;
1093 };
1094
1095 my $get_container_namespace = sub {
1096     my ($vmid, $pid, $kind) = @_;
1097
1098     my $pidfd;
1099     if (!defined($pid)) {
1100         # Pin the pid while we're grabbing its stuff from /proc
1101         ($pid, $pidfd) = open_lxc_pid($vmid)
1102             or die "failed to open pidfd of container $vmid\'s init process\n";
1103     }
1104
1105     return $open_namespace->($vmid, $pid, $kind);
1106 };
1107
1108 my $do_syncfs = sub {
1109     my ($vmid, $pid, $socket) = @_;
1110
1111     &$enter_namespace($vmid, $pid, 'mnt', PVE::Tools::CLONE_NEWNS);
1112
1113     # Tell the parent process to start reading our /proc/mounts
1114     print {$socket} "go\n";
1115     $socket->flush();
1116
1117     # Receive /proc/self/mounts
1118     my $mountdata = do { local $/ = undef; <$socket> };
1119     close $socket;
1120
1121     my %nosyncfs = (
1122         cgroup => 1,
1123         cgroup2 => 1,
1124         devtmpfs => 1,
1125         devpts => 1,
1126         'fuse.lxcfs' => 1,
1127         fusectl => 1,
1128         mqueue => 1,
1129         proc => 1,
1130         sysfs => 1,
1131         tmpfs => 1,
1132     );
1133
1134     # Now sync all mountpoints...
1135     my $mounts = PVE::ProcFSTools::parse_mounts($mountdata);
1136     foreach my $mp (@$mounts) {
1137         my ($what, $dir, $fs) = @$mp;
1138         next if $nosyncfs{$fs};
1139         eval { PVE::Tools::sync_mountpoint($dir); };
1140         warn $@ if $@;
1141     }
1142 };
1143
1144 sub sync_container_namespace {
1145     my ($vmid) = @_;
1146     my $pid = find_lxc_pid($vmid);
1147
1148     # SOCK_DGRAM is nicer for barriers but cannot be slurped
1149     socketpair my $pfd, my $cfd, AF_UNIX, SOCK_STREAM, PF_UNSPEC
1150         or die "failed to create socketpair: $!\n";
1151
1152     my $child = fork();
1153     die "fork failed: $!\n" if !defined($child);
1154
1155     if (!$child) {
1156         eval {
1157             close $pfd;
1158             &$do_syncfs($vmid, $pid, $cfd);
1159         };
1160         if (my $err = $@) {
1161             warn $err;
1162             POSIX::_exit(1);
1163         }
1164         POSIX::_exit(0);
1165     }
1166     close $cfd;
1167     my $go = <$pfd>;
1168     die "failed to enter container namespace\n" if $go ne "go\n";
1169
1170     open my $mounts, '<', "/proc/$child/mounts"
1171         or die "failed to open container's /proc/mounts: $!\n";
1172     my $mountdata = do { local $/ = undef; <$mounts> };
1173     close $mounts;
1174     print {$pfd} $mountdata;
1175     close $pfd;
1176
1177     while (waitpid($child, 0) != $child) {}
1178     die "failed to sync container namespace\n" if $? != 0;
1179 }
1180
1181 sub template_create {
1182     my ($vmid, $conf) = @_;
1183
1184     my $storecfg = PVE::Storage::config();
1185
1186     PVE::LXC::Config->foreach_volume($conf, sub {
1187         my ($ms, $mountpoint) = @_;
1188
1189         my $volid = $mountpoint->{volume};
1190
1191         die "Template feature is not available for '$volid'\n"
1192             if !PVE::Storage::volume_has_feature($storecfg, 'template', $volid);
1193     });
1194
1195     PVE::LXC::Config->foreach_volume($conf, sub {
1196         my ($ms, $mountpoint) = @_;
1197
1198         my $volid = $mountpoint->{volume};
1199
1200         PVE::Storage::activate_volumes($storecfg, [$volid]);
1201
1202         my $template_volid = PVE::Storage::vdisk_create_base($storecfg, $volid);
1203         $mountpoint->{volume} = $template_volid;
1204         $conf->{$ms} = PVE::LXC::Config->print_ct_mountpoint($mountpoint, $ms eq "rootfs");
1205     });
1206
1207     PVE::LXC::Config->write_config($vmid, $conf);
1208 }
1209
1210 sub check_ct_modify_config_perm {
1211     my ($rpcenv, $authuser, $vmid, $pool, $newconf, $delete) = @_;
1212
1213     return 1 if $authuser eq 'root@pam';
1214
1215     my $check = sub {
1216         my ($opt, $delete) = @_;
1217         if ($opt eq 'cores' || $opt eq 'cpuunits' || $opt eq 'cpulimit') {
1218             $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.CPU']);
1219         } elsif ($opt eq 'rootfs' || $opt =~ /^mp\d+$/) {
1220             $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.Disk']);
1221             return if $delete;
1222             my $data = PVE::LXC::Config->parse_volume($opt, $newconf->{$opt});
1223             raise_perm_exc("mount point type $data->{type} is only allowed for root\@pam")
1224                 if $data->{type} ne 'volume';
1225         } elsif ($opt eq 'memory' || $opt eq 'swap') {
1226             $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.Memory']);
1227         } elsif ($opt =~ m/^net\d+$/ || $opt eq 'nameserver' ||
1228                  $opt eq 'searchdomain' || $opt eq 'hostname') {
1229             $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.Network']);
1230         } elsif ($opt eq 'features') {
1231             # For now this is restricted to root@pam
1232             raise_perm_exc("changing feature flags is only allowed for root\@pam");
1233         } elsif ($opt eq 'hookscript') {
1234             # For now this is restricted to root@pam
1235             raise_perm_exc("changing the hookscript is only allowed for root\@pam");
1236         } else {
1237             $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.Options']);
1238         }
1239     };
1240
1241     foreach my $opt (keys %$newconf) {
1242         &$check($opt, 0);
1243     }
1244     foreach my $opt (@$delete) {
1245         &$check($opt, 1);
1246     }
1247
1248     return 1;
1249 }
1250
1251 sub umount_all {
1252     my ($vmid, $storage_cfg, $conf, $noerr) = @_;
1253
1254     my $rootdir = "/var/lib/lxc/$vmid/rootfs";
1255     my $volid_list = PVE::LXC::Config->get_vm_volumes($conf);
1256
1257     my $res = 1;
1258
1259     PVE::LXC::Config->foreach_volume_full($conf, {'reverse' => 1}, sub {
1260         my ($ms, $mountpoint) = @_;
1261
1262         my $volid = $mountpoint->{volume};
1263         my $mount = $mountpoint->{mp};
1264
1265         return if !$volid || !$mount;
1266
1267         my $mount_path = "$rootdir/$mount";
1268         $mount_path =~ s!/+!/!g;
1269
1270         return if !PVE::ProcFSTools::is_mounted($mount_path);
1271
1272         eval {
1273             PVE::Tools::run_command(['umount', '-d', $mount_path]);
1274         };
1275         if (my $err = $@) {
1276             if ($noerr) {
1277                 $res = 0;
1278                 warn $err;
1279             } else {
1280                 die $err;
1281             }
1282         }
1283     });
1284
1285     return $res; # tell caller if (some) umounts failed for the noerr case
1286 }
1287
1288 sub mount_all {
1289     my ($vmid, $storage_cfg, $conf, $ignore_ro) = @_;
1290
1291     my $rootdir = "/var/lib/lxc/$vmid/rootfs";
1292     File::Path::make_path($rootdir);
1293
1294     my $volid_list = PVE::LXC::Config->get_vm_volumes($conf);
1295     PVE::Storage::activate_volumes($storage_cfg, $volid_list);
1296
1297     my (undef, $rootuid, $rootgid) = parse_id_maps($conf);
1298
1299     eval {
1300         PVE::LXC::Config->foreach_volume($conf, sub {
1301             my ($ms, $mountpoint) = @_;
1302
1303             $mountpoint->{ro} = 0 if $ignore_ro;
1304
1305             mountpoint_mount($mountpoint, $rootdir, $storage_cfg, undef, $rootuid, $rootgid);
1306         });
1307     };
1308     if (my $err = $@) {
1309         warn "mounting container failed\n";
1310         umount_all($vmid, $storage_cfg, $conf, 1);
1311         die $err;
1312     }
1313
1314     return $rootdir;
1315 }
1316
1317
1318 sub mountpoint_mount_path {
1319     my ($mountpoint, $storage_cfg, $snapname) = @_;
1320
1321     return mountpoint_mount($mountpoint, undef, $storage_cfg, $snapname);
1322 }
1323
1324 sub query_loopdev {
1325     my ($path) = @_;
1326     my $found;
1327     my $parser = sub {
1328         my $line = shift;
1329         if ($line =~ m@^(/dev/loop\d+):@) {
1330             $found = $1;
1331         }
1332     };
1333     my $cmd = ['losetup', '--associated', $path];
1334     PVE::Tools::run_command($cmd, outfunc => $parser);
1335     return $found;
1336 }
1337
1338 # Run a function with a file attached to a loop device.
1339 # The loop device is always detached afterwards (or set to autoclear).
1340 # Returns the loop device.
1341 sub run_with_loopdev {
1342     my ($func, $file, $readonly) = @_;
1343     my $device = query_loopdev($file);
1344     # Try to reuse an existing device
1345     if ($device) {
1346         # We assume that whoever setup the loop device is responsible for
1347         # detaching it.
1348         &$func($device);
1349         return $device;
1350     }
1351
1352     my $parser = sub {
1353         my $line = shift;
1354         if ($line =~ m@^(/dev/loop\d+)$@) {
1355             $device = $1;
1356         }
1357     };
1358     my $losetup_cmd = [
1359         'losetup',
1360         '--show',
1361         '-f',
1362         $file,
1363     ];
1364     push @$losetup_cmd, '-r' if $readonly;
1365     PVE::Tools::run_command($losetup_cmd, outfunc => $parser);
1366     die "failed to setup loop device for $file\n" if !$device;
1367     eval { &$func($device); };
1368     my $err = $@;
1369     PVE::Tools::run_command(['losetup', '-d', $device]);
1370     die $err if $err;
1371     return $device;
1372 }
1373
1374 # In scalar mode: returns a file handle to the deepest directory node.
1375 # In list context: returns a list of:
1376 #   * the deepest directory node
1377 #   * the 2nd deepest directory (parent of the above)
1378 #   * directory name of the last directory
1379 # So that the path $2/$3 should lead to $1 afterwards.
1380 sub walk_tree_nofollow($$$;$$) {
1381     my ($start, $subdir, $mkdir, $rootuid, $rootgid) = @_;
1382
1383     sysopen(my $fd, $start, O_PATH | O_DIRECTORY)
1384         or die "failed to open start directory $start: $!\n";
1385
1386     return walk_tree_nofollow_fd($start, $fd, $subdir, $mkdir, $rootuid, $rootgid);
1387 }
1388
1389
1390 sub walk_tree_nofollow_fd($$$$;$$) {
1391     my ($start_dirname, $start_fd, $subdir, $mkdir, $rootuid, $rootgid) = @_;
1392
1393     # splitdir() returns '' for empty components including the leading /
1394     my @comps = grep { length($_)>0 } File::Spec->splitdir($subdir);
1395
1396     my $fd = $start_fd;
1397     my $dir = $start_dirname;
1398     my $last_component = undef;
1399     my $second = $fd;
1400     foreach my $component (@comps) {
1401         $dir .= "/$component";
1402         my $next = PVE::Tools::openat(fileno($fd), $component, O_NOFOLLOW | O_DIRECTORY);
1403
1404         if (!$next) {
1405             # failed, check for symlinks and try to create the path
1406             die "symlink encountered at: $dir\n" if $! == ELOOP || $! == ENOTDIR;
1407             die "cannot open directory $dir: $!\n" if !$mkdir;
1408
1409             # We don't check for errors on mkdirat() here and just try to
1410             # openat() again, since at least one error (EEXIST) is an
1411             # expected possibility if multiple containers start
1412             # simultaneously. If someone else injects a symlink now then
1413             # the subsequent openat() will fail due to O_NOFOLLOW anyway.
1414             PVE::Tools::mkdirat(fileno($fd), $component, 0755);
1415
1416             $next = PVE::Tools::openat(fileno($fd), $component, O_NOFOLLOW | O_DIRECTORY);
1417             die "failed to create path: $dir: $!\n" if !$next;
1418
1419             PVE::Tools::fchownat(fileno($next), '', $rootuid, $rootgid, PVE::Tools::AT_EMPTY_PATH)
1420                 if defined($rootuid) && defined($rootgid);
1421         }
1422
1423         close $second if defined($last_component) && $second != $start_fd;
1424         $last_component = $component;
1425         $second = $fd;
1426         $fd = $next;
1427     }
1428
1429     return ($fd, defined($last_component) && $second, $last_component) if wantarray;
1430     close $second if defined($last_component) && $second != $start_fd;
1431     return $fd;
1432 }
1433
1434 # To guard against symlink attack races against other currently running
1435 # containers with shared recursive bind mount hierarchies we prepare a
1436 # directory handle for the directory we're mounting over to verify the
1437 # mountpoint afterwards.
1438 sub __bindmount_prepare {
1439     my ($hostroot, $dir) = @_;
1440     my $srcdh = walk_tree_nofollow($hostroot, $dir, 0);
1441     return $srcdh;
1442 }
1443
1444 # Assuming we mount to rootfs/a/b/c, verify with the directory handle to 'b'
1445 # ($parentfd) that 'b/c' (openat($parentfd, 'c')) really leads to the directory
1446 # we intended to bind mount.
1447 sub __bindmount_verify {
1448     my ($srcdh, $parentfd, $last_dir, $ro) = @_;
1449     my $destdh;
1450     if ($parentfd) {
1451         # Open the mount point path coming from the parent directory since the
1452         # filehandle we would have gotten as first result of walk_tree_nofollow
1453         # earlier is still a handle to the underlying directory instead of the
1454         # mounted path.
1455         $destdh = PVE::Tools::openat(fileno($parentfd), $last_dir, PVE::Tools::O_PATH | O_NOFOLLOW | O_DIRECTORY);
1456         die "failed to open mount point: $!\n" if !$destdh;
1457         if ($ro) {
1458             my $dot = '.';
1459             # no separate function because 99% of the time it's the wrong thing to use.
1460             if (syscall(PVE::Syscall::faccessat, fileno($destdh), $dot, &POSIX::W_OK, 0) != -1) {
1461                 die "failed to mark bind mount read only\n";
1462             }
1463             die "read-only check failed: $!\n" if $! != EROFS;
1464         }
1465     } else {
1466         # For the rootfs we don't have a parentfd so we open the path directly.
1467         # Note that this means bindmounting any prefix of the host's
1468         # /var/lib/lxc/$vmid path into another container is considered a grave
1469         # security error.
1470         sysopen $destdh, $last_dir, O_PATH | O_DIRECTORY;
1471         die "failed to open mount point: $!\n" if !$destdh;
1472     }
1473
1474     my ($srcdev, $srcinode) = stat($srcdh);
1475     my ($dstdev, $dstinode) = stat($destdh);
1476     close $srcdh;
1477     close $destdh;
1478
1479     return ($srcdev == $dstdev && $srcinode == $dstinode);
1480 }
1481
1482 # Perform the actual bind mounting:
1483 sub __bindmount_do {
1484     my ($dir, $dest, $ro, @extra_opts) = @_;
1485     PVE::Tools::run_command(['mount', '-o', 'bind', @extra_opts, $dir, $dest]);
1486     if ($ro) {
1487         eval { PVE::Tools::run_command(['mount', '-o', 'bind,remount,ro', $dest]); };
1488         if (my $err = $@) {
1489             warn "bindmount error\n";
1490             # don't leave writable bind-mounts behind...
1491             PVE::Tools::run_command(['umount', $dest]);
1492             die $err;
1493         }
1494     }
1495 }
1496
1497 sub bindmount {
1498     my ($dir, $parentfd, $last_dir, $dest, $ro, @extra_opts) = @_;
1499
1500     my $srcdh = __bindmount_prepare('/', $dir);
1501
1502     __bindmount_do($dir, $dest, $ro, @extra_opts);
1503
1504     if (!__bindmount_verify($srcdh, $parentfd, $last_dir, $ro)) {
1505         PVE::Tools::run_command(['umount', $dest]);
1506         die "detected mount path change at: $dir\n";
1507     }
1508 }
1509
1510 # Cleanup $rootdir a bit (double and trailing slashes), build the mount path
1511 # from $rootdir and $mount and walk the path from $rootdir to the final
1512 # directory to check for symlinks.
1513 sub __mount_prepare_rootdir {
1514     my ($rootdir, $mount, $rootuid, $rootgid) = @_;
1515     $rootdir =~ s!/+!/!g;
1516     $rootdir =~ s!/+$!!;
1517     my $mount_path = "$rootdir/$mount";
1518     my ($mpfd, $parentfd, $last_dir) = walk_tree_nofollow($rootdir, $mount, 1, $rootuid, $rootgid);
1519     return ($rootdir, $mount_path, $mpfd, $parentfd, $last_dir);
1520 }
1521
1522 # use $rootdir = undef to just return the corresponding mount path
1523 sub mountpoint_mount {
1524     my ($mountpoint, $rootdir, $storage_cfg, $snapname, $rootuid, $rootgid) = @_;
1525     return __mountpoint_mount($mountpoint, $rootdir, $storage_cfg, $snapname, $rootuid, $rootgid, undef);
1526 }
1527
1528 sub mountpoint_stage {
1529     my ($mountpoint, $stage_dir, $storage_cfg, $snapname, $rootuid, $rootgid) = @_;
1530     my ($path, $loop, $dev) =
1531         __mountpoint_mount($mountpoint, $stage_dir, $storage_cfg, $snapname, $rootuid, $rootgid, 1);
1532
1533     if (!defined($path)) {
1534         return undef if $! == ENOSYS;
1535         die "failed to mount subvolume: $!\n";
1536     }
1537
1538     # We clone the mount point and leave it there in order to keep them connected to eg. loop
1539     # devices in case we're hotplugging (which would allow contaienrs to unmount the new mount
1540     # point).
1541     my $err;
1542     my $fd = PVE::Tools::open_tree(&AT_FDCWD, $stage_dir, &OPEN_TREE_CLOEXEC | &OPEN_TREE_CLONE)
1543         or die "open_tree() on mount point failed: $!\n";
1544
1545     return wantarray ? ($path, $loop, $dev, $fd) : $fd;
1546 }
1547
1548 sub mountpoint_insert_staged {
1549     my ($mount_fd, $rootdir_fd, $mp_dir, $opt, $rootuid, $rootgid) = @_;
1550
1551     if (!defined($rootdir_fd)) {
1552         sysopen($rootdir_fd, '.', O_PATH | O_DIRECTORY)
1553             or die "failed to open '.': $!\n";
1554     }
1555
1556     my $dest_fd = walk_tree_nofollow_fd('/', $rootdir_fd, $mp_dir, 1, $rootuid, $rootgid);
1557
1558     PVE::Tools::move_mount(
1559         fileno($mount_fd),
1560         '',
1561         fileno($dest_fd),
1562         '',
1563         &MOVE_MOUNT_F_EMPTY_PATH | &MOVE_MOUNT_T_EMPTY_PATH,
1564     ) or die "failed to move '$opt' into container hierarchy: $!\n";
1565 }
1566
1567 # Use $stage_mount, $rootdir is treated as a temporary path to "stage" the file system. The user
1568 #   can then open a file descriptor to it which can be used with the `move_mount` syscall.
1569 #   Note that if the kernel does not support the new mount API, this will not perform any action
1570 #   and return `undef` with $! = ENOSYS.
1571 sub __mountpoint_mount {
1572     my ($mountpoint, $rootdir, $storage_cfg, $snapname, $rootuid, $rootgid, $stage_mount) = @_;
1573
1574     if (defined($stage_mount) && !PVE::LXC::Tools::can_use_new_mount_api()) {
1575         $! = ENOSYS;
1576         return undef;
1577     }
1578
1579     # When staging mount points we always mount to $rootdir directly (iow. as if `mp=/`).
1580     # This is required since __mount_prepare_rootdir() will return handles to the parent directory
1581     # which we use in __bindmount_verify()!
1582     my $mount = $stage_mount ? '/': $mountpoint->{mp};
1583
1584     my $volid = $mountpoint->{volume};
1585     my $type = $mountpoint->{type};
1586     my $quota = !$snapname && !$mountpoint->{ro} && $mountpoint->{quota};
1587     my $mounted_dev;
1588
1589     return if !$volid || !$mount;
1590
1591     $mount =~ s!/+!/!g;
1592
1593     my $mount_path;
1594     my ($mpfd, $parentfd, $last_dir);
1595
1596     if (defined($rootdir)) {
1597         ($rootdir, $mount_path, $mpfd, $parentfd, $last_dir) =
1598             __mount_prepare_rootdir($rootdir, $mount, $rootuid, $rootgid);
1599     }
1600
1601     if (defined($stage_mount)) {
1602         $mount_path = $rootdir;
1603     }
1604
1605     my ($storage, $volname) = PVE::Storage::parse_volume_id($volid, 1);
1606
1607     die "unknown snapshot path for '$volid'" if !$storage && defined($snapname);
1608
1609     my $optlist = [];
1610
1611     if (my $mountopts = $mountpoint->{mountoptions}) {
1612         my @opts = split(/;/, $mountpoint->{mountoptions});
1613         push @$optlist, grep { PVE::LXC::Config::is_valid_mount_option($_) } @opts;
1614     }
1615
1616     my $acl = $mountpoint->{acl};
1617     if (defined($acl)) {
1618         push @$optlist, ($acl ? 'acl' : 'noacl');
1619     }
1620
1621     my $optstring = join(',', @$optlist);
1622     my $readonly = $mountpoint->{ro};
1623
1624     my @extra_opts;
1625     @extra_opts = ('-o', $optstring) if $optstring;
1626
1627     if ($storage) {
1628
1629         my $scfg = PVE::Storage::storage_config($storage_cfg, $storage);
1630
1631         my $path = PVE::Storage::map_volume($storage_cfg, $volid, $snapname);
1632
1633         $path = PVE::Storage::path($storage_cfg, $volid, $snapname) if !defined($path);
1634
1635         my ($vtype, undef, undef, undef, undef, $isBase, $format) =
1636             PVE::Storage::parse_volname($storage_cfg, $volid);
1637
1638         $format = 'iso' if $vtype eq 'iso'; # allow to handle iso files
1639
1640         if ($format eq 'subvol') {
1641             if ($mount_path) {
1642                 my (undef, $name) = PVE::Storage::parse_volname($storage_cfg, $volid);
1643                 if (defined($snapname)) {
1644                     $name .= "\@$snapname";
1645                     if ($scfg->{type} eq 'zfspool') {
1646                         PVE::Tools::run_command(['mount', '-o', 'ro', @extra_opts, '-t', 'zfs', "$scfg->{pool}/$name", $mount_path]);
1647                     } else {
1648                         die "cannot mount subvol snapshots for storage type '$scfg->{type}'\n";
1649                     }
1650                 } else {
1651                     if (defined($acl) && $scfg->{type} eq 'zfspool') {
1652                         my $acltype = ($acl ? 'acltype=posixacl' : 'acltype=noacl');
1653                         PVE::Tools::run_command(['zfs', 'set', $acltype, "$scfg->{pool}/$name"]);
1654                     }
1655                     bindmount($path, $parentfd, $last_dir//$rootdir, $mount_path, $readonly, @extra_opts);
1656                     warn "cannot enable quota control for bind mounted subvolumes\n" if $quota;
1657                 }
1658             }
1659             return wantarray ? ($path, 0, undef) : $path;
1660         } elsif ($format eq 'raw' || $format eq 'iso') {
1661             # NOTE: 'mount' performs canonicalization without the '-c' switch, which for
1662             # device-mapper devices is special-cased to use the /dev/mapper symlinks.
1663             # Our autodev hook expects the /dev/dm-* device currently
1664             # and will create the /dev/mapper symlink accordingly
1665             $path = Cwd::realpath($path);
1666             die "failed to get device path\n" if !$path;
1667             ($path) = ($path =~ /^(.*)$/s); #untaint
1668             my $domount = sub {
1669                 my ($path) = @_;
1670                 if ($mount_path) {
1671                     if ($format eq 'iso') {
1672                         PVE::Tools::run_command(['mount', '-o', 'ro', @extra_opts, $path, $mount_path]);
1673                     } elsif ($isBase || defined($snapname)) {
1674                         PVE::Tools::run_command(['mount', '-o', 'ro,noload', @extra_opts, $path, $mount_path]);
1675                     } else {
1676                         if ($quota) {
1677                             push @extra_opts, '-o', 'usrjquota=aquota.user,grpjquota=aquota.group,jqfmt=vfsv0';
1678                         }
1679                         push @extra_opts, '-o', 'ro' if $readonly;
1680                         PVE::Tools::run_command(['mount', @extra_opts, $path, $mount_path]);
1681                     }
1682                 }
1683             };
1684             my $use_loopdev = 0;
1685             if ($scfg->{path}) {
1686                 $mounted_dev = run_with_loopdev($domount, $path, $readonly);
1687                 $use_loopdev = 1;
1688             } elsif ($scfg->{type} eq 'drbd' || $scfg->{type} eq 'lvm' ||
1689                      $scfg->{type} eq 'rbd' || $scfg->{type} eq 'lvmthin') {
1690                 $mounted_dev = $path;
1691                 &$domount($path);
1692             } else {
1693                 die "unsupported storage type '$scfg->{type}'\n";
1694             }
1695             return wantarray ? ($path, $use_loopdev, $mounted_dev) : $path;
1696         } else {
1697             die "unsupported image format '$format'\n";
1698         }
1699     } elsif ($type eq 'device') {
1700         push @extra_opts, '-o', 'ro' if $readonly;
1701         push @extra_opts, '-o', 'usrjquota=aquota.user,grpjquota=aquota.group,jqfmt=vfsv0' if $quota;
1702         # See the NOTE above about devicemapper canonicalization
1703         my ($devpath) = (Cwd::realpath($volid) =~ /^(.*)$/s); # realpath() taints
1704         PVE::Tools::run_command(['mount', @extra_opts, $volid, $mount_path]) if $mount_path;
1705         return wantarray ? ($volid, 0, $devpath) : $volid;
1706     } elsif ($type eq 'bind') {
1707         die "directory '$volid' does not exist\n" if ! -d $volid;
1708         bindmount($volid, $parentfd, $last_dir//$rootdir, $mount_path, $readonly, @extra_opts) if $mount_path;
1709         warn "cannot enable quota control for bind mounts\n" if $quota;
1710         return wantarray ? ($volid, 0, undef) : $volid;
1711     }
1712
1713     die "unsupported storage";
1714 }
1715
1716 sub mountpoint_hotplug($$$) {
1717     my ($vmid, $conf, $opt, $mp, $storage_cfg) = @_;
1718
1719     my (undef, $rootuid, $rootgid) = PVE::LXC::parse_id_maps($conf);
1720
1721     # We do the rest in a fork with an unshared mount namespace, because:
1722     #  -) change our papparmor profile to that of /usr/bin/lxc-start
1723     #  -) we're now going to 'stage' # the mountpoint, then grab it, then move into the
1724     #     container's namespace, then mount it.
1725
1726     PVE::Tools::run_fork(sub {
1727         # Pin the container pid longer, we also need to get its monitor/parent:
1728         my ($ct_pid, $ct_pidfd) = open_lxc_pid($vmid)
1729             or die "failed to open pidfd of container $vmid\'s init process\n";
1730
1731         my ($monitor_pid, $monitor_pidfd) = open_ppid($ct_pid)
1732             or die "failed to open pidfd of container $vmid\'s monitor process\n";
1733
1734         my $ct_mnt_ns = $get_container_namespace->($vmid, $ct_pid, 'mnt');
1735         my $monitor_mnt_ns = $get_container_namespace->($vmid, $monitor_pid, 'mnt');
1736
1737         # Grab a file descriptor to our apparmor label file so we can change into the 'lxc-start'
1738         # profile to lower our privileges to the same level we have in the start hook:
1739         sysopen(my $aa_fd, "/proc/self/attr/current", O_WRONLY)
1740             or die "failed to open '/proc/self/attr/current' for writing: $!\n";
1741         # But switch namespaces first, to make sure the namespace switches aren't blocked by
1742         # apparmor.
1743
1744         # Change into the monitor's mount namespace. We "pin" the mount into the monitor's
1745         # namespace for it to remain active there since the container will be able to unmount
1746         # hotplugged mount points and thereby potentially free up loop devices, which is a security
1747         # concern.
1748         PVE::Tools::setns(fileno($monitor_mnt_ns), PVE::Tools::CLONE_NEWNS);
1749         chdir('/')
1750             or die "failed to change root directory within the monitor's mount namespace: $!\n";
1751
1752         my $dir = get_staging_mount_path($opt);
1753
1754         # Now switch our apparmor profile before mounting:
1755         my $data = 'changeprofile /usr/bin/lxc-start';
1756         if (syswrite($aa_fd, $data, length($data)) != length($data)) {
1757             die "failed to change apparmor profile: $!\n";
1758         }
1759         # Check errors on close as well:
1760         close($aa_fd)
1761             or die "failed to change apparmor profile (close() failed): $!\n";
1762
1763         my $mount_fd = mountpoint_stage($mp, $dir, $storage_cfg, undef, $rootuid, $rootgid);
1764
1765         PVE::Tools::setns(fileno($ct_mnt_ns), PVE::Tools::CLONE_NEWNS);
1766         chdir('/')
1767             or die "failed to change root directory within the container's mount namespace: $!\n";
1768
1769         mountpoint_insert_staged($mount_fd, undef, $mp->{mp}, $opt, $rootuid, $rootgid);
1770     });
1771 }
1772
1773 # Create a directory in the mountpoint staging tempfs.
1774 sub get_staging_mount_path($) {
1775     my ($opt) = @_;
1776
1777     my $target = get_staging_tempfs() . "/$opt";
1778     if (!mkdir($target) && $! != EEXIST) {
1779         die "failed to create directory $target: $!\n";
1780     }
1781
1782     return $target;
1783 }
1784
1785 # Mount /run/pve/mountpoints as tmpfs
1786 sub get_staging_tempfs() {
1787     # We choose a path in /var/lib/lxc/ here because the lxc-start apparmor profile restricts most
1788     # mounts to that.
1789     my $target = '/var/lib/lxc/.pve-staged-mounts';
1790     if (!mkdir($target)) {
1791         return $target if $! == EEXIST;
1792         die "failed to create directory $target: $!\n";
1793     }
1794
1795     PVE::Tools::mount("none", $target, 'tmpfs', 0, "size=8k,mode=755")
1796         or die "failed to mount $target as tmpfs: $!\n";
1797
1798     return $target;
1799 }
1800
1801 sub mkfs {
1802     my ($dev, $rootuid, $rootgid) = @_;
1803
1804     run_command(
1805         [
1806             'mkfs.ext4',
1807             '-O',
1808             'mmp',
1809             '-E',
1810             "root_owner=$rootuid:$rootgid",
1811             $dev,
1812         ],
1813         outfunc => sub {
1814             my $line = shift;
1815             # a hack to print only the relevant stuff, i.e., the one which could help on repair
1816             if ($line =~ /^(Creating filesystem|Filesystem UUID|Superblock backups|\s+\d+, \d)/) {
1817                 print "$line\n";
1818             }
1819         },
1820         errfunc => sub {
1821             my $line = shift;
1822             print STDERR "$line\n" if $line && $line !~ /^mke2fs \d\.\d/;
1823         }
1824     );
1825 }
1826
1827 sub format_disk {
1828     my ($storage_cfg, $volid, $rootuid, $rootgid) = @_;
1829
1830     if ($volid =~ m!^/dev/.+!) {
1831         mkfs($volid);
1832         return;
1833     }
1834
1835     my ($storage, $volname) = PVE::Storage::parse_volume_id($volid, 1);
1836
1837     die "cannot format volume '$volid' with no storage\n" if !$storage;
1838
1839     PVE::Storage::activate_volumes($storage_cfg, [$volid]);
1840
1841     my $path = PVE::Storage::map_volume($storage_cfg, $volid);
1842
1843     $path = PVE::Storage::path($storage_cfg, $volid) if !defined($path);
1844
1845     my ($vtype, undef, undef, undef, undef, $isBase, $format) =
1846         PVE::Storage::parse_volname($storage_cfg, $volid);
1847
1848     die "cannot format volume '$volid' (format == $format)\n"
1849         if $format ne 'raw';
1850
1851     mkfs($path, $rootuid, $rootgid);
1852 }
1853
1854 sub destroy_disks {
1855     my ($storecfg, $vollist) = @_;
1856
1857     foreach my $volid (@$vollist) {
1858         eval { PVE::Storage::vdisk_free($storecfg, $volid); };
1859         warn $@ if $@;
1860     }
1861 }
1862
1863 sub alloc_disk {
1864     my ($storecfg, $vmid, $storage, $size_kb, $rootuid, $rootgid) = @_;
1865
1866     my $needs_chown = 0;
1867     my $volid;
1868
1869     my $scfg = PVE::Storage::storage_config($storecfg, $storage);
1870     # fixme: use better naming ct-$vmid-disk-X.raw?
1871
1872     eval {
1873         my $do_format = 0;
1874         if ($scfg->{type} eq 'dir' || $scfg->{type} eq 'nfs' || $scfg->{type} eq 'cifs' ) {
1875             if ($size_kb > 0) {
1876                 $volid = PVE::Storage::vdisk_alloc($storecfg, $storage, $vmid, 'raw',
1877                                                    undef, $size_kb);
1878                 $do_format = 1;
1879             } else {
1880                 $volid = PVE::Storage::vdisk_alloc($storecfg, $storage, $vmid, 'subvol',
1881                                                    undef, 0);
1882                 $needs_chown = 1;
1883             }
1884         } elsif ($scfg->{type} eq 'zfspool') {
1885
1886             $volid = PVE::Storage::vdisk_alloc($storecfg, $storage, $vmid, 'subvol',
1887                                                undef, $size_kb);
1888             $needs_chown = 1;
1889         } elsif ($scfg->{type} eq 'drbd' || $scfg->{type} eq 'lvm' || $scfg->{type} eq 'lvmthin') {
1890
1891             $volid = PVE::Storage::vdisk_alloc($storecfg, $storage, $vmid, 'raw', undef, $size_kb);
1892             $do_format = 1;
1893
1894         } elsif ($scfg->{type} eq 'rbd') {
1895
1896             $volid = PVE::Storage::vdisk_alloc($storecfg, $storage, $vmid, 'raw', undef, $size_kb);
1897             $do_format = 1;
1898         } else {
1899             die "unable to create containers on storage type '$scfg->{type}'\n";
1900         }
1901         format_disk($storecfg, $volid, $rootuid, $rootgid) if $do_format;
1902     };
1903     if (my $err = $@) {
1904         # in case formatting got interrupted:
1905         if (defined($volid)) {
1906             eval { PVE::Storage::vdisk_free($storecfg, $volid); };
1907             warn $@ if $@;
1908         }
1909         die $err;
1910     }
1911
1912     return ($volid, $needs_chown);
1913 }
1914
1915 our $NEW_DISK_RE = qr/^([^:\s]+):(\d+(\.\d+)?)$/;
1916 sub create_disks {
1917     my ($storecfg, $vmid, $settings, $conf, $pending) = @_;
1918
1919     my $vollist = [];
1920
1921     eval {
1922         my (undef, $rootuid, $rootgid) = PVE::LXC::parse_id_maps($conf);
1923         my $chown_vollist = [];
1924
1925         PVE::LXC::Config->foreach_volume($settings, sub {
1926             my ($ms, $mountpoint) = @_;
1927
1928             my $volid = $mountpoint->{volume};
1929             my $mp = $mountpoint->{mp};
1930
1931             my ($storage, $volname) = PVE::Storage::parse_volume_id($volid, 1);
1932
1933             if ($storage && ($volid =~ $NEW_DISK_RE)) {
1934                 my ($storeid, $size_gb) = ($1, $2);
1935
1936                 my $size_kb = int(${size_gb}*1024) * 1024;
1937
1938                 my $needs_chown = 0;
1939                 ($volid, $needs_chown) = alloc_disk($storecfg, $vmid, $storage, $size_kb, $rootuid, $rootgid);
1940                 push @$chown_vollist, $volid if $needs_chown;
1941                 push @$vollist, $volid;
1942                 $mountpoint->{volume} = $volid;
1943                 $mountpoint->{size} = $size_kb * 1024;
1944                 if ($pending) {
1945                     $conf->{pending}->{$ms} = PVE::LXC::Config->print_ct_mountpoint($mountpoint, $ms eq 'rootfs');
1946                 } else {
1947                     $conf->{$ms} = PVE::LXC::Config->print_ct_mountpoint($mountpoint, $ms eq 'rootfs');
1948                 }
1949             } else {
1950                 # use specified/existing volid/dir/device
1951                 $conf->{$ms} = PVE::LXC::Config->print_ct_mountpoint($mountpoint, $ms eq 'rootfs');
1952             }
1953         });
1954
1955         PVE::Storage::activate_volumes($storecfg, $chown_vollist, undef);
1956         foreach my $volid (@$chown_vollist) {
1957             my $path = PVE::Storage::path($storecfg, $volid, undef);
1958             chown($rootuid, $rootgid, $path);
1959         }
1960         PVE::Storage::deactivate_volumes($storecfg, $chown_vollist, undef);
1961     };
1962     # free allocated images on error
1963     if (my $err = $@) {
1964         destroy_disks($storecfg, $vollist);
1965         die $err;
1966     }
1967     return $vollist;
1968 }
1969
1970 sub update_disksize {
1971     my ($vmid, $conf, $all_volumes) = @_;
1972
1973     my $changes;
1974     my $prefix = "CT $vmid:";
1975
1976     my $update_mp = sub {
1977         my ($key, $mp, @param) = @_;
1978         my $size = $all_volumes->{$mp->{volume}}->{size} // 0;
1979
1980         if (!defined($mp->{size}) || $size != $mp->{size}) {
1981             $changes = 1;
1982             print "$prefix updated volume size of '$mp->{volume}' in config.\n";
1983             $mp->{size} = $size;
1984             my $nomp = 1 if ($key eq 'rootfs');
1985             $conf->{$key} = PVE::LXC::Config->print_ct_mountpoint($mp, $nomp);
1986         }
1987     };
1988
1989     PVE::LXC::Config->foreach_volume($conf, $update_mp);
1990
1991     return $changes;
1992 }
1993
1994 sub update_unused {
1995     my ($vmid, $conf, $all_volumes) = @_;
1996
1997     my $changes;
1998     my $prefix = "CT $vmid:";
1999
2000     # Note: it is allowed to define multiple storage entries with the same path
2001     # (alias), so we need to check both 'volid' and real 'path' (two different
2002     # volid can point to the same path).
2003
2004     # used and unused disks
2005     my $refpath = {};
2006     my $orphans = {};
2007
2008     foreach my $opt (keys %$conf) {
2009         next if ($opt !~ m/^unused\d+$/);
2010         my $vol = $all_volumes->{$conf->{$opt}};
2011         $refpath->{$vol->{path}} = $vol->{volid};
2012     }
2013
2014     foreach my $key (keys %$all_volumes) {
2015         my $vol = $all_volumes->{$key};
2016         my $in_use = PVE::LXC::Config->is_volume_in_use($conf, $vol->{volid});
2017         my $path = $vol->{path};
2018
2019         if ($in_use) {
2020             $refpath->{$path} = $key;
2021             delete $orphans->{$path};
2022         } else {
2023             if ((!$orphans->{$path}) && (!$refpath->{$path})) {
2024                 $orphans->{$path} = $key;
2025             }
2026         }
2027     }
2028
2029     for my $key (keys %$orphans) {
2030         my $disk = $orphans->{$key};
2031         my $unused = PVE::LXC::Config->add_unused_volume($conf, $disk);
2032
2033         if ($unused) {
2034             $changes = 1;
2035             print "$prefix add unreferenced volume '$disk' as '$unused' to config.\n";
2036         }
2037     }
2038
2039     return $changes;
2040 }
2041
2042 sub scan_volids {
2043     my ($cfg, $vmid) = @_;
2044
2045     my $info = PVE::Storage::vdisk_list($cfg, undef, $vmid, undef, 'rootdir');
2046
2047     my $all_volumes = {};
2048     foreach my $storeid (keys %$info) {
2049         foreach my $item (@{$info->{$storeid}}) {
2050             my $volid = $item->{volid};
2051             next if !($volid && $item->{size});
2052             $item->{path} = PVE::Storage::path($cfg, $volid);
2053             $all_volumes->{$volid} = $item;
2054         }
2055     }
2056
2057     return $all_volumes;
2058 }
2059
2060 sub rescan {
2061     my ($vmid, $nolock, $dryrun) = @_;
2062
2063     my $cfg = PVE::Storage::config();
2064
2065     print "rescan volumes...\n";
2066     my $all_volumes = scan_volids($cfg, $vmid);
2067
2068     my $updatefn =  sub {
2069         my ($vmid) = @_;
2070
2071         my $changes;
2072         my $conf = PVE::LXC::Config->load_config($vmid);
2073
2074         PVE::LXC::Config->check_lock($conf);
2075
2076         my $vm_volids = {};
2077         foreach my $volid (keys %$all_volumes) {
2078             my $info = $all_volumes->{$volid};
2079             $vm_volids->{$volid} = $info if $info->{vmid} == $vmid;
2080         }
2081
2082         my $upu = update_unused($vmid, $conf, $vm_volids);
2083         my $upd = update_disksize($vmid, $conf, $vm_volids);
2084         $changes = $upu || $upd;
2085
2086         PVE::LXC::Config->write_config($vmid, $conf) if $changes && !$dryrun;
2087     };
2088
2089     if (defined($vmid)) {
2090         if ($nolock) {
2091             &$updatefn($vmid);
2092         } else {
2093             PVE::LXC::Config->lock_config($vmid, $updatefn, $vmid);
2094         }
2095     } else {
2096         my $vmlist = config_list();
2097         foreach my $vmid (keys %$vmlist) {
2098             if ($nolock) {
2099                 &$updatefn($vmid);
2100             } else {
2101                 PVE::LXC::Config->lock_config($vmid, $updatefn, $vmid);
2102             }
2103         }
2104     }
2105 }
2106
2107
2108 # bash completion helper
2109
2110 sub complete_os_templates {
2111     my ($cmdname, $pname, $cvalue) = @_;
2112
2113     my $cfg = PVE::Storage::config();
2114
2115     my $storeid;
2116
2117     if ($cvalue =~ m/^([^:]+):/) {
2118         $storeid = $1;
2119     }
2120
2121     my $vtype = $cmdname eq 'restore' ? 'backup' : 'vztmpl';
2122     my $data = PVE::Storage::template_list($cfg, $storeid, $vtype);
2123
2124     my $res = [];
2125     foreach my $id (keys %$data) {
2126         foreach my $item (@{$data->{$id}}) {
2127             push @$res, $item->{volid} if defined($item->{volid});
2128         }
2129     }
2130
2131     return $res;
2132 }
2133
2134 my $complete_ctid_full = sub {
2135     my ($running) = @_;
2136
2137     my $idlist = vmstatus();
2138
2139     my $active_hash = list_active_containers();
2140
2141     my $res = [];
2142
2143     foreach my $id (keys %$idlist) {
2144         my $d = $idlist->{$id};
2145         if (defined($running)) {
2146             next if $d->{template};
2147             next if $running && !$active_hash->{$id};
2148             next if !$running && $active_hash->{$id};
2149         }
2150         push @$res, $id;
2151
2152     }
2153     return $res;
2154 };
2155
2156 sub complete_ctid {
2157     return &$complete_ctid_full();
2158 }
2159
2160 sub complete_ctid_stopped {
2161     return &$complete_ctid_full(0);
2162 }
2163
2164 sub complete_ctid_running {
2165     return &$complete_ctid_full(1);
2166 }
2167
2168 sub parse_id_maps {
2169     my ($conf) = @_;
2170
2171     my $id_map = [];
2172     my $rootuid = 0;
2173     my $rootgid = 0;
2174
2175     my $lxc = $conf->{lxc};
2176     foreach my $entry (@$lxc) {
2177         my ($key, $value) = @$entry;
2178         # FIXME: remove the 'id_map' variant when lxc-3.0 arrives
2179         next if $key ne 'lxc.idmap' && $key ne 'lxc.id_map';
2180         if ($value =~ /^([ug])\s+(\d+)\s+(\d+)\s+(\d+)\s*$/) {
2181             my ($type, $ct, $host, $length) = ($1, $2, $3, $4);
2182             push @$id_map, [$type, $ct, $host, $length];
2183             if ($ct == 0) {
2184                 $rootuid = $host if $type eq 'u';
2185                 $rootgid = $host if $type eq 'g';
2186             }
2187         } else {
2188             die "failed to parse idmap: $value\n";
2189         }
2190     }
2191
2192     if (!@$id_map && $conf->{unprivileged}) {
2193         # Should we read them from /etc/subuid?
2194         $id_map = [ ['u', '0', '100000', '65536'],
2195                     ['g', '0', '100000', '65536'] ];
2196         $rootuid = $rootgid = 100000;
2197     }
2198
2199     return ($id_map, $rootuid, $rootgid);
2200 }
2201
2202 sub userns_command {
2203     my ($id_map) = @_;
2204     if (@$id_map) {
2205         return ['lxc-usernsexec', (map { ('-m', join(':', @$_)) } @$id_map), '--'];
2206     }
2207     return [];
2208 }
2209
2210 my sub print_ct_stderr_log {
2211     my ($vmid) = @_;
2212     my $log = eval { file_get_contents("/run/pve/ct-$vmid.stderr") };
2213     return if !$log;
2214
2215     while ($log =~ /^\h*(lxc-start:?\s+$vmid:?\s*\S+\s*)?(.*?)\h*$/gm) {
2216         my $line = $2;
2217         print STDERR "$line\n";
2218     }
2219 }
2220
2221 my sub monitor_state_change($$) {
2222     my ($monitor_socket, $vmid) = @_;
2223     die "no monitor socket\n" if !defined($monitor_socket);
2224
2225     while (1) {
2226         my ($type, $name, $value) = PVE::LXC::Monitor::read_lxc_message($monitor_socket);
2227
2228         die "monitor socket: got EOF\n" if !defined($type);
2229
2230         next if $name ne "$vmid" || $type ne 'STATE';
2231
2232         if ($value eq PVE::LXC::Monitor::STATE_STARTING) {
2233             alarm(0); # don't timeout after seeing the starting state
2234         } elsif ($value eq PVE::LXC::Monitor::STATE_ABORTING ||
2235                  $value eq PVE::LXC::Monitor::STATE_STOPPING ||
2236                  $value eq PVE::LXC::Monitor::STATE_STOPPED) {
2237             return 0;
2238         } elsif ($value eq PVE::LXC::Monitor::STATE_RUNNING) {
2239             return 1;
2240         } else {
2241             warn "unexpected message from monitor socket - " .
2242                  "type: '$type' - value: '$value'\n";
2243         }
2244     }
2245 }
2246 my sub monitor_start($$) {
2247     my ($monitor_socket, $vmid) = @_;
2248
2249     my $success = eval {
2250         PVE::Tools::run_with_timeout(10, \&monitor_state_change, $monitor_socket, $vmid)
2251     };
2252     if (my $err = $@) {
2253         warn "problem with monitor socket, but continuing anyway: $err\n";
2254     } elsif (!$success) {
2255         print_ct_stderr_log($vmid);
2256         die "startup for container '$vmid' failed\n";
2257     }
2258 }
2259
2260 sub vm_start {
2261     my ($vmid, $conf, $skiplock, $debug) = @_;
2262
2263     # apply pending changes while starting
2264     if (scalar(keys %{$conf->{pending}})) {
2265         my $storecfg = PVE::Storage::config();
2266         PVE::LXC::Config->vmconfig_apply_pending($vmid, $conf, $storecfg);
2267         PVE::LXC::Config->write_config($vmid, $conf);
2268         $conf = PVE::LXC::Config->load_config($vmid); # update/reload
2269     }
2270
2271     update_lxc_config($vmid, $conf);
2272
2273     my $skiplock_flag_fn = "/run/lxc/skiplock-$vmid";
2274
2275     if ($skiplock) {
2276         open(my $fh, '>', $skiplock_flag_fn) || die "failed to open $skiplock_flag_fn for writing: $!\n";
2277         close($fh);
2278     }
2279
2280     my $storage_cfg = PVE::Storage::config();
2281     my $vollist = PVE::LXC::Config->get_vm_volumes($conf);
2282
2283     PVE::Storage::activate_volumes($storage_cfg, $vollist);
2284
2285     my $monitor_socket = eval { PVE::LXC::Monitor::get_monitor_socket() };
2286     warn $@ if $@;
2287
2288     unlink "/run/pve/ct-$vmid.stderr"; # systemd does not truncate log files
2289
2290     my $is_debug = $debug || (!defined($debug) && $conf->{debug});
2291     my $base_unit = $is_debug ? 'pve-container-debug' : 'pve-container';
2292
2293     my $cmd = ['systemctl', 'start', "$base_unit\@$vmid"];
2294
2295     PVE::GuestHelpers::exec_hookscript($conf, $vmid, 'pre-start', 1);
2296     eval {
2297         run_command($cmd);
2298
2299         monitor_start($monitor_socket, $vmid) if defined($monitor_socket);
2300
2301         # if debug is requested, print the log it also when the start succeeded
2302         print_ct_stderr_log($vmid) if $is_debug;
2303     };
2304     if (my $err = $@) {
2305         unlink $skiplock_flag_fn;
2306         die $err;
2307     }
2308     PVE::GuestHelpers::exec_hookscript($conf, $vmid, 'post-start');
2309
2310     return;
2311 }
2312
2313 # Helper to stop a container completely and make sure it has stopped completely.
2314 # This is necessary because we want the post-stop hook to have completed its
2315 # unmount-all step, but post-stop happens after lxc puts the container into the
2316 # STOPPED state.
2317 # $kill - if true it will always do an immediate hard-stop
2318 # $shutdown_timeout - the timeout to wait for a gracefull shutdown
2319 # $kill_after_timeout - if true, send a hardstop if shutdown timed out
2320 sub vm_stop {
2321     my ($vmid, $kill, $shutdown_timeout, $kill_after_timeout) = @_;
2322
2323     # Open the container's command socket.
2324     my $path = "\0/var/lib/lxc/$vmid/command";
2325     my $sock = IO::Socket::UNIX->new(
2326         Type => SOCK_STREAM(),
2327         Peer => $path,
2328     );
2329     if (!$sock) {
2330         return if $! == ECONNREFUSED; # The container is not running
2331         die "failed to open container ${vmid}'s command socket: $!\n";
2332     }
2333
2334     my $conf = PVE::LXC::Config->load_config($vmid);
2335     PVE::GuestHelpers::exec_hookscript($conf, $vmid, 'pre-stop');
2336
2337     # Stop the container:
2338
2339     my $cmd = ['lxc-stop', '-n', $vmid];
2340
2341     if ($kill) {
2342         push @$cmd, '--kill'; # doesn't allow timeouts
2343     } else {
2344         # lxc-stop uses a default timeout
2345         push @$cmd, '--nokill' if !$kill_after_timeout;
2346
2347         if (defined($shutdown_timeout)) {
2348             push @$cmd, '--timeout', $shutdown_timeout;
2349             # Give run_command 5 extra seconds
2350             $shutdown_timeout += 5;
2351         }
2352     }
2353
2354     eval { run_command($cmd, timeout => $shutdown_timeout) };
2355     if (my $err = $@) {
2356         warn $@ if $@;
2357     }
2358
2359     my $result = <$sock>;
2360
2361     return if !defined $result; # monitor is gone and the ct has stopped.
2362     die "container did not stop\n";
2363 }
2364
2365 sub vm_reboot {
2366     my ($vmid, $timeout, $skiplock) = @_;
2367
2368     PVE::LXC::Config->lock_config($vmid, sub {
2369         return if !check_running($vmid);
2370
2371         vm_stop($vmid, 0, $timeout, 1); # kill if timeout exceeds
2372
2373         my $conf = PVE::LXC::Config->load_config($vmid);
2374         vm_start($vmid, $conf);
2375     });
2376 }
2377
2378 sub run_unshared {
2379     my ($code) = @_;
2380
2381     return PVE::Tools::run_fork(sub {
2382         # Unshare the mount namespace
2383         die "failed to unshare mount namespace: $!\n"
2384             if !PVE::Tools::unshare(PVE::Tools::CLONE_NEWNS);
2385         run_command(['mount', '--make-rslave', '/']);
2386         return $code->();
2387     });
2388 }
2389
2390 my $copy_volume = sub {
2391     my ($src_volid, $src, $dst_volid, $dest, $storage_cfg, $snapname, $bwlimit, $rootuid,  $rootgid) = @_;
2392
2393     my $src_mp = { volume => $src_volid, mp => '/', ro => 1 };
2394     $src_mp->{type} = PVE::LXC::Config->classify_mountpoint($src_volid);
2395
2396     my $dst_mp = { volume => $dst_volid, mp => '/', ro => 0 };
2397     $dst_mp->{type} = PVE::LXC::Config->classify_mountpoint($dst_volid);
2398
2399     my @mounted;
2400     eval {
2401         # mount and copy
2402         mkdir $src;
2403         mountpoint_mount($src_mp, $src, $storage_cfg, $snapname, $rootuid, $rootgid);
2404         push @mounted, $src;
2405         mkdir $dest;
2406         mountpoint_mount($dst_mp, $dest, $storage_cfg, undef, $rootuid, $rootgid);
2407         push @mounted, $dest;
2408
2409         $bwlimit //= 0;
2410
2411         run_command([
2412             'rsync',
2413             '--stats',
2414             '-X',
2415             '-A',
2416             '--numeric-ids',
2417             '-aH',
2418             '--whole-file',
2419             '--sparse',
2420             '--one-file-system',
2421             "--bwlimit=$bwlimit",
2422             "$src/",
2423             $dest
2424         ]);
2425     };
2426     my $err = $@;
2427
2428     # Wait for rsync's children to release dest so that
2429     # consequent file operations (umount, remove) are possible
2430     while ((system {"fuser"} "fuser",  "-s", $dest) == 0) {sleep 1};
2431
2432     foreach my $mount (reverse @mounted) {
2433         eval { run_command(['/bin/umount', $mount], errfunc => sub{})};
2434         warn "Can't umount $mount\n" if $@;
2435     }
2436
2437     # If this fails they're used as mount points in a concurrent operation
2438     # (which should not happen but there's also no real need to get rid of them).
2439     rmdir $dest;
2440     rmdir $src;
2441
2442     die $err if $err;
2443 };
2444
2445 # Should not be called after unsharing the mount namespace!
2446 sub copy_volume {
2447     my ($mp, $vmid, $storage, $storage_cfg, $conf, $snapname, $bwlimit) = @_;
2448
2449     die "cannot copy volumes of type $mp->{type}\n" if $mp->{type} ne 'volume';
2450     File::Path::make_path("/var/lib/lxc/$vmid");
2451     my $dest = "/var/lib/lxc/$vmid/.copy-volume-1";
2452     my $src  = "/var/lib/lxc/$vmid/.copy-volume-2";
2453
2454     # get id's for unprivileged container
2455     my (undef, $rootuid, $rootgid) = parse_id_maps($conf);
2456
2457     # Allocate the disk before unsharing in order to make sure zfs subvolumes
2458     # are visible in this namespace, otherwise the host only sees the empty
2459     # (not-mounted) directory.
2460     my $new_volid;
2461     eval {
2462         # Make sure $mp contains a correct size.
2463         $mp->{size} = PVE::Storage::volume_size_info($storage_cfg, $mp->{volume});
2464         my $needs_chown;
2465         ($new_volid, $needs_chown) = alloc_disk($storage_cfg, $vmid, $storage, $mp->{size}/1024, $rootuid, $rootgid);
2466         if ($needs_chown) {
2467             PVE::Storage::activate_volumes($storage_cfg, [$new_volid], undef);
2468             my $path = PVE::Storage::path($storage_cfg, $new_volid, undef);
2469             chown($rootuid, $rootgid, $path);
2470         }
2471
2472         run_unshared(sub {
2473             $copy_volume->($mp->{volume}, $src, $new_volid, $dest, $storage_cfg, $snapname, $bwlimit, $rootuid, $rootgid);
2474         });
2475     };
2476     if (my $err = $@) {
2477         PVE::Storage::vdisk_free($storage_cfg, $new_volid)
2478             if defined($new_volid);
2479         die $err;
2480     }
2481
2482     return $new_volid;
2483 }
2484
2485 sub get_lxc_version() {
2486     my $version;
2487     run_command([qw(lxc-start --version)], outfunc => sub {
2488         my ($line) = @_;
2489         # We only parse out major & minor version numbers.
2490         if ($line =~ /^(\d+)\.(\d+)(?:\D.*)?$/) {
2491             $version = [$1, $2];
2492         }
2493     });
2494
2495     die "failed to get lxc version\n" if !defined($version);
2496
2497     # return as a list:
2498     return $version->@*;
2499 }
2500
2501 sub freeze($) {
2502     my ($vmid) = @_;
2503     if (PVE::CGroup::cgroup_mode() == 2) {
2504         PVE::LXC::Command::freeze($vmid, 30);
2505     } else {
2506         PVE::LXC::CGroup->new($vmid)->freeze_thaw(1);
2507     }
2508 }
2509
2510 sub thaw($) {
2511     my ($vmid) = @_;
2512     if (PVE::CGroup::cgroup_mode() == 2) {
2513         PVE::LXC::Command::unfreeze($vmid, 30);
2514     } else {
2515         PVE::LXC::CGroup->new($vmid)->freeze_thaw(0);
2516     }
2517 }
2518
2519 1;